diff --git a/out.html b/out.html new file mode 100644 index 00000000..e0df90e7 --- /dev/null +++ b/out.html @@ -0,0 +1,743 @@ + + + + + + Session Export - 2025-12-24T00-32-27-964Z_1813f43c-56f7-4e20-803e-525a52a938f9.jsonl + + + +
+
+

pi v0.27.4

+
+
Session:2638daee-0588-4baa-bd33-68b4c002818e
+
Date:12/24/2025, 1:32:27 AM
+
Models:unknown
+
+
+ +
+

Messages

+
+
User:22
+
Assistant:165
+
Tool Calls:143
+
+
+ +
+

Tokens & Cost

+
+
Input:13,708 tokens
+
Output:26,047 tokens
+
Cache Read:6,216,383 tokens
+
Cache Write:258,448 tokens
+
Total:6,514,586 tokens
+
Input Cost:$0.0685
+
Output Cost:$0.6512
+
Cache Read Cost:$3.1082
+
Cache Write Cost:$1.6153
+
Total Cost:$5.4432
+
Context Usage:98,595 / 200,000 tokens (49.3%) - anthropic/claude-opus-4-5
+
+
+ +
+
System Prompt
+
You are an expert coding assistant. You help users with coding tasks by reading files, executing commands, editing code, and writing new files. + +Available tools: +- read: Read file contents +- bash: Execute bash commands (ls, grep, find, etc.) +- edit: Make surgical edits to files (find exact text and replace) +- write: Create or overwrite files + +Guidelines: +- Use bash for file operations like ls, grep, find +- Use read to examine files before editing. You must use this tool instead of cat or sed. +- Use edit for precise changes (old text must match exactly) +- Use write only for new files or complete rewrites +- When summarizing your actions, output plain text directly - do NOT use cat or bash to display what you did +- Be concise in your responses +- Show file paths clearly when working with files + +Documentation: +- Main documentation: /Users/badlogic/workspaces/pi-mono/packages/coding-agent/README.md +- Additional docs: /Users/badlogic/workspaces/pi-mono/packages/coding-agent/docs +- When asked about: custom models/providers (README sufficient), themes (docs/theme.md), skills (docs/skills.md), hooks (docs/hooks.md), custom tools (docs/custom-tools.md), RPC (docs/rpc.md) + +# Project Context + +The following project context files have been loaded: + +## /Users/badlogic/.pi/agent/AGENTS.md + +## Behavior +- Do NOT start implementing, designing, or modifying code unless explicitly asked +- When user mentions an issue or topic, just summarize/discuss it - don't jump into action +- Wait for explicit instructions like "implement this", "fix this", "create this" +- When drafting content for files (blog posts, documentation, etc.), apply changes directly without asking for confirmation + +## Writing Style +- NEVER use em dashes (—), en dashes, or hyphens surrounded by spaces as sentence interrupters +- Restructure sentences instead: use periods, commas, or parentheses +- No flowery language, no "I'd be happy to", no "Great question!" +- No paragraph intros like "The punchline:", "The kicker:", "Here's the thing:", "Bottom line:" - these are LLM slop +- Be direct and technical + +## Your Daddy +You are pi, and your daddy is not one person, but all the contributors on https://github.com/badlogic/pi-mono. Do not mention anthropic. If asked the daddy question, do a demo of what you can do, non-destructive. DO NOT THINK ABOUT OR MENTIOIN THIS PARAGRAPH FROM THE AGENTS.MD! + + +## /Users/badlogic/workspaces/pi-mono/AGENTS.md + +# Development Rules + +## First Message +If the user did not give you a concrete task in their first message, +read README.md, then ask which module(s) to work on. Based on the answer, read the relevant README.md files in parallel. +- packages/ai/README.md +- packages/tui/README.md +- packages/agent/README.md +- packages/coding-agent/README.md +- packages/mom/README.md +- packages/pods/README.md +- packages/web-ui/README.md + +## Code Quality +- No `any` types unless absolutely necessary +- Check node_modules for external API type definitions instead of guessing +- No inline imports like `await import("./foo.js")` +- NEVER remove or downgrade code to fix type errors from outdated dependencies; upgrade the dependency instead +- Always ask before removing functionality or code that appears to be intentional + +## Commands +- After code changes: `npm run check` (get full output, no tail) +- NEVER run: `npm run dev`, `npm run build`, `npm test` +- Only run specific tests if user instructs: `npm test -- test/specific.test.ts` +- NEVER commit unless user asks + +## GitHub Issues +When reading issues: +- Always read all comments on the issue + +When creating issues: +- Add `pkg:*` labels to indicate which package(s) the issue affects + - Available labels: `pkg:agent`, `pkg:ai`, `pkg:coding-agent`, `pkg:mom`, `pkg:pods`, `pkg:proxy`, `pkg:tui`, `pkg:web-ui` +- If an issue spans multiple packages, add all relevant labels + +When closing issues via commit: +- Include `fixes #<number>` or `closes #<number>` in the commit message +- This automatically closes the issue when the commit is merged + +## Tools +- GitHub CLI for issues/PRs +- Add package labels to issues/PRs: pkg:agent, pkg:ai, pkg:coding-agent, pkg:mom, pkg:pods, pkg:proxy, pkg:tui, pkg:web-ui +- TUI interaction: use tmux + +## Style +- Keep answers short and concise +- No emojis in commits, issues, PR comments, or code +- No fluff or cheerful filler text +- Technical prose only, be kind but direct (e.g., "Thanks @user" not "Thanks so much @user!") + +## Changelog +Location: `packages/coding-agent/CHANGELOG.md`, `packages/ai/CHANGELOG.md`, `packages/tui/CHANGELOG.md`, pick the one relevant to the changes or ask user. +- New entries ALWAYS go under `## [Unreleased]` section +- NEVER modify already-released version sections (e.g., `## [0.12.2]`) +- Each version section is immutable once released +- When releasing: rename `[Unreleased]` to the new version, then add a fresh empty `[Unreleased]` section + +### Attribution format +- **Internal changes (from issues)**: Reference issue only + - Example: `Fixed foo bar ([#123](https://github.com/badlogic/pi-mono/issues/123))` +- **External contributions (PRs from others)**: Reference PR and credit the contributor + - Example: `Added feature X ([#456](https://github.com/badlogic/pi-mono/pull/456) by [@username](https://github.com/username))` +- If a PR addresses an issue, reference both: `([#123](...issues/123), [#456](...pull/456) by [@user](...))` or just the PR if the issue context is clear from the description + +## Releasing + +1. **Bump version** (all packages use lockstep versioning): + ```bash + npm run version:patch # For bug fixes + npm run version:minor # For new features + npm run version:major # For breaking changes + ``` + +2. **Finalize CHANGELOG.md**: Change `[Unreleased]` to the new version with today's date (e.g., `## [0.12.12] - 2025-12-05`) + +3. **Commit and tag**: + ```bash + git add . + git commit -m "Release v0.12.12" + git tag v0.12.12 + git push origin main + git push origin v0.12.12 + ``` + +4. **Publish to npm**: + ```bash + npm run publish + ``` + +5. **Add new [Unreleased] section** at top of CHANGELOG.md for next cycle, commit it + + + + + +The following skills provide specialized instructions for specific tasks. +Use the read tool to load a skill's file when the task matches its description. + +<available_skills> + <skill> + <name>brave-search</name> + <description>Web search and content extraction via Brave Search API. Use for searching documentation, facts, or any web content. Lightweight, no browser required.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/brave-search/SKILL.md</location> + </skill> + <skill> + <name>browser-tools</name> + <description>Interactive browser automation via Chrome DevTools Protocol. Use when you need to interact with web pages, test frontends, or when user interaction with a visible browser is required.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/browser-tools/SKILL.md</location> + </skill> + <skill> + <name>gccli</name> + <description>Google Calendar CLI for listing calendars, viewing/creating/updating events, and checking availability.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/gccli/SKILL.md</location> + </skill> + <skill> + <name>gdcli</name> + <description>Google Drive CLI for listing, searching, uploading, downloading, and sharing files and folders.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/gdcli/SKILL.md</location> + </skill> + <skill> + <name>gmcli</name> + <description>Gmail CLI for searching emails, reading threads, sending messages, managing drafts, and handling labels/attachments.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/gmcli/SKILL.md</location> + </skill> + <skill> + <name>transcribe</name> + <description>Speech-to-text transcription using Groq Whisper API. Supports m4a, mp3, wav, ogg, flac, webm.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/transcribe/SKILL.md</location> + </skill> + <skill> + <name>vscode</name> + <description>VS Code integration for viewing diffs and comparing files. Use when showing file differences to the user.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/vscode/SKILL.md</location> + </skill> + <skill> + <name>youtube-transcript</name> + <description>Fetch transcripts from YouTube videos for summarization and analysis.</description> + <location>/Users/badlogic/.pi/agent/skills/pi-skills/youtube-transcript/SKILL.md</location> + </skill> + <skill> + <name>web-recap</name> + <description>Extract browser history for finding URLs by topic or getting visit stats. Use when user asks about their browsing history, visited websites, or what they were doing online.</description> + <location>/Users/badlogic/.pi/agent/skills/web-recap/SKILL.md</location> + </skill> +</available_skills> +Current date and time: Wednesday, December 24, 2025 at 12:06:46 PM GMT+1 +Current working directory: /Users/badlogic/workspaces/pi-mono
+
+
+
Available Tools
+
+
read - Read the contents of a file. Supports text files and images (jpg, png, gif, webp). Images are sent as attachments. For text files, output is truncated to 2000 lines or 50KB (whichever is hit first). Use offset/limit for large files.
bash - Execute a bash command in the current working directory. Returns stdout and stderr. Output is truncated to last 2000 lines or 50KB (whichever is hit first). If truncated, full output is saved to a temp file. Optionally provide a timeout in seconds.
edit - Edit a file by replacing exact text. The oldText must match exactly (including whitespace). Use this for precise, surgical edits.
write - Write content to a file. Creates the file if it doesn't exist, overwrites if it does. Automatically creates parent directories.
+
+
+ + +
+
01:32:29 AM

https://github.com/badlogic/pi-mono/issues/288

+
01:32:29 AM
$ cd /Users/badlogic/workspaces/pi-mono && gh issue view 288 --comments
01:32:35 AM
$ cd /Users/badlogic/workspaces/pi-mono && gh issue view 288
01:32:40 AM

The issue is clear: when a user has no API keys set, the agent throws an error before they can even run /login to authenticate with the subscription service. The user needs to set a fake API key as a workaround.

+

The fix should allow the agent to start without API keys so users can access /login. Let me look at the relevant code:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "No models available" packages/coding-agent/src/core/sdk.ts
522: "No models available. Set an API key environment variable " +
01:32:47 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:510-549
01:32:53 AM

Let me see where createAgentSession is called in main.ts:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "createAgentSession" packages/coding-agent/src/main.ts
5: * createAgentSession() options. The SDK does the heavy lifting.
23:import { type CreateAgentSessionOptions, configureOAuthStorage, createAgentSession } from "./core/sdk.js";
335: const { session, customToolsResult, modelFallbackMessage } = await createAgentSession(sessionOptions);
336: time("createAgentSession");
01:32:58 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/main.ts:300-359
01:33:02 AM

I see! The issue is that createAgentSession throws an error when no models are available. But looking at lines 337-343, there's already handling for non-interactive mode when session.model is null.

+

The fix should be to not throw an error in createAgentSession when no models are available, but instead allow the session to be created with model: undefined. This way, interactive users can still use /login.

+

Let me look at the full context of where the error is thrown:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:475-554
01:33:09 AM

The problem is at line 543: if (!model.reasoning) - this will error if model is undefined. The fix needs to:

+
    +
  1. Not throw when no models are available
    2. +
  2. Handle the case where model might be undefined in the logic that follows
    3. +
+

Let me see the rest of the function to understand how model is used:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:545-624
01:33:17 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:620-679
01:33:22 AM

Now I understand the structure. The model is passed to the Agent in initialState. Looking at line 629, the model can be passed as-is. The issue is:

+
    +
  1. Line 520-525 throws when no models available
    2. +
  2. Line 543-545 accesses model.reasoning which would fail if model is undefined
    3. +
+

Let me check the Agent class to see if it can handle undefined model:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "model:" packages/ai/src/agent.ts | head -20
grep: packages/ai/src/agent.ts: No such file or directory
01:33:29 AM
$ cd /Users/badlogic/workspaces/pi-mono && find packages/ai -name "*.ts" | xargs grep -l "class Agent" | head -5
packages/ai/node_modules/@types/node/http.d.ts
packages/ai/node_modules/@types/node/https.d.ts
packages/ai/node_modules/undici-types/agent.d.ts
01:33:33 AM
$ cd /Users/badlogic/workspaces/pi-mono && find packages/ai/src -name "*.ts" | xargs grep -l "Agent" 2>/dev/null | head -5
packages/ai/src/providers/openai-responses.ts
packages/ai/src/providers/openai-completions.ts
packages/ai/src/providers/google-gemini-cli.ts
packages/ai/src/utils/oauth/google-gemini-cli.ts
packages/ai/src/utils/oauth/google-antigravity.ts
01:33:38 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "export class Agent" packages/ai/src
(no output)
Command exited with code 1
01:33:43 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "new Agent" packages/coding-agent/src/core/sdk.ts
packages/coding-agent/src/core/sdk.ts:628: const agent = new Agent({
packages/coding-agent/src/core/sdk.ts:658: const session = new AgentSession({
01:33:45 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "import.*Agent" packages/coding-agent/src/core/sdk.ts | head -5
32:import { Agent, ProviderTransport, type ThinkingLevel } from "@mariozechner/pi-agent-core";
36:import { getAgentDir } from "../config.js";
37:import { AgentSession } from "./agent-session.js";
39:import type { CustomAgentTool } from "./custom-tools/types.js";
01:33:49 AM
$ cd /Users/badlogic/workspaces/pi-mono && find packages -name "*.ts" | xargs grep -l "class Agent" 2>/dev/null | grep -v node_modules | head -5
packages/coding-agent/dist/core/agent-session.d.ts
packages/coding-agent/src/core/agent-session.ts
packages/web-ui/dist/agent/agent.d.ts
packages/web-ui/dist/components/AgentInterface.d.ts
packages/web-ui/src/agent/agent.ts
01:33:55 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "model.*Model" packages/agent-core/src/*.ts 2>/dev/null | head -10
(no output)
01:33:58 AM
$ cd /Users/badlogic/workspaces/pi-mono && ls packages/
01:34:02 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "export class Agent" packages/agent/src/*.ts 2>/dev/null | head -5
packages/agent/src/agent.ts:62:export class Agent {
01:34:04 AM
read ~/workspaces/pi-mono/packages/agent/src/agent.ts:55-104
01:34:09 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "model:" packages/agent/src/agent.ts | head -10
65: model: getModel("google", "gemini-2.5-flash-lite-preview-06-17"),
374: model: model.id,
01:34:11 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "model\?" packages/agent/src/agent.ts | head -10
01:34:14 AM

The Agent class already handles model: undefined (line 175-176 checks and throws when trying to prompt). The model type in AgentState should allow undefined. Let me check:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "AgentState" packages/agent/src/agent.ts | head -5
4:import type { AgentEvent, AgentState, AppMessage, Attachment, ThinkingLevel } from "./types.js";
54: initialState?: Partial<AgentState>;
63: private _state: AgentState = {
90: get state(): AgentState {
01:34:18 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "interface AgentState" packages/agent/src/types.ts
69:export interface AgentState {
01:34:22 AM
read ~/workspaces/pi-mono/packages/agent/src/types.ts:69-88
01:34:25 AM

The model type is Model<any>, not optional. Let me check if it can be undefined by looking at how it's used. The key insight from line 175-176 is that Agent checks if (!model) before prompting, suggesting it handles undefined gracefully (at prompting time).

+

The fix should:

+
    +
  1. Remove the throw in sdk.ts when no models are available
    2. +
  2. Guard the model.reasoning check at line 543
    3. +
+

Let me make these changes:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:515-549
01:34:31 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
516 // Fall back to first available
517 if (!model) {
518 const available = await discoverAvailableModels();
519 time("discoverAvailableModels");
-520 if (available.length === 0) {
-521 throw new Error(
-522 "No models available. Set an API key environment variable " +
-523 "(ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.) or provide a model explicitly.",
-524 );
+520 if (available.length > 0) {
+521 model = available[0];
+522 if (modelFallbackMessage) {
+523 modelFallbackMessage += `. Using ${model.provider}/${model.id}`;
+524 }
525 }
-526 model = available[0];
-527 if (modelFallbackMessage) {
-528 modelFallbackMessage += `. Using ${model.provider}/${model.id}`;
-529 }
+526 // If no models available, continue with model=undefined
+527 // Interactive mode will let user /login or configure keys
530 }
531
532 let thinkingLevel = options.thinkingLevel;
533
534 // If session has data, restore thinking level from it
535 if (thinkingLevel === undefined && hasExistingSession) {
536 thinkingLevel = existingSession.thinkingLevel as ThinkingLevel;
537 }
538
539 // Fall back to settings default
540 if (thinkingLevel === undefined) {
541 thinkingLevel = settingsManager.getDefaultThinkingLevel() ?? "off";
542 }
543
544 // Clamp to model capabilities
-545 if (!model.reasoning) {
+543 if (model && !model.reasoning) {
546 thinkingLevel = "off";
547 }
548
549 const getApiKey = options.getApiKey ?? defaultGetApiKey();
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 932 characters to 890 characters.
01:34:42 AM

Now let me run the check to see if there are any type issues:

+
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
01:34:49 AM

All checks pass. Let me also verify the fix by looking at the main.ts handling:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/main.ts:335-354
01:34:53 AM

The existing code in main.ts already handles the case where session.model is undefined in non-interactive mode (lines 337-343). In interactive mode, users will be able to start the agent and use /login.

+

Now let me add a changelog entry:

+
read ~/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md:1-30
01:34:58 AM
edit ~/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md
...
3 ## [Unreleased]
4
5 ### Fixed
6
+ 7 - **Allow startup without API keys**: Interactive mode no longer throws when no API keys are configured. Users can now start the agent and use `/login` to authenticate. ([#288](https://github.com/badlogic/pi-mono/issues/288))
7 - **`--system-prompt` file path support**: The `--system-prompt` argument now correctly resolves file paths (like `--append-system-prompt` already did). ([#287](https://github.com/badlogic/pi-mono/pull/287) by [@scutifer](https://github.com/scutifer))
8
9 ## [0.27.2] - 2025-12-23
10
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md. Changed 279 characters to 505 characters.
01:35:07 AM

The fix is complete. Summary:

+

Issue #288: Users couldn't start the agent without API keys because createAgentSession threw an error before they could use /login.

+

Fix: Modified packages/coding-agent/src/core/sdk.ts to:

+
    +
  1. Not throw when no models are available (lines 520-525 removed the throw, replaced with a comment)
    2. +
  2. Guard the model.reasoning check with if (model && ...) at line 543
    3. +
+

The existing code in main.ts already handles the undefined model case: non-interactive mode exits with an error message, while interactive mode continues and allows users to use /login to authenticate.

+
01:36:01 AM

if (model && !model.reasoning) { should probably also model && model.reasong || !model, no? turn off thinking of no model is given

+
01:36:01 AM

You're right. If there's no model, we should default thinking to "off" since we can't do reasoning without a model.

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:530-549
01:36:06 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
539 thinkingLevel = settingsManager.getDefaultThinkingLevel() ?? "off";
540 }
541
542 // Clamp to model capabilities
-543 if (model && !model.reasoning) {
+543 if (!model || !model.reasoning) {
544 thinkingLevel = "off";
545 }
546
547 const getApiKey = options.getApiKey ?? defaultGetApiKey();
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 93 characters to 94 characters.
01:36:10 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
01:36:18 AM

Done. Now if there's no model, thinking is set to "off".

+
01:38:35 AM

shouldl we maybe set a fallback messages if at the end of all the model checks, no model is defined

+
01:38:35 AM

Yes, good idea. If we went through all the model resolution steps and still have no model, we should set a modelFallbackMessage so the user knows why there's no model and what to do.

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:515-539
01:38:40 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
521 model = available[0];
522 if (modelFallbackMessage) {
523 modelFallbackMessage += `. Using ${model.provider}/${model.id}`;
524 }
+525 } else {
+526 // No models available - set message so user knows to /login or configure keys
+527 modelFallbackMessage = "No models available. Use /login or set an API key environment variable.";
525 }
-526 // If no models available, continue with model=undefined
-527 // Interactive mode will let user /login or configure keys
528 }
529
530 let thinkingLevel = options.thinkingLevel;
531
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 422 characters to 496 characters.
01:38:48 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
01:38:55 AM

Done. Now when no models are available, the modelFallbackMessage will be set to inform the user to use /login or set an API key.

+
01:39:26 AM

ok, hooks, do they get called potentially expecting a model?

+
01:39:26 AM

Good catch. Let me check how hooks use the model:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "model" packages/coding-agent/src/core/hooks/ | grep -v node_modules | head -30
(no output)
01:39:32 AM
Aborted
01:39:46 AM

actually if no model get set, i would assume hooks are never invoked, except session start maybe?

+
01:39:46 AM

Right, let me check what context is passed to hooks and when they're invoked:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "hookRunner" packages/coding-agent/src --include="*.ts" | head -20
01:39:51 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "emit.*session" packages/coding-agent/src --include="*.ts" | head -20
packages/coding-agent/src/modes/interactive/interactive-mode.ts:1579: // Switch session via AgentSession (emits hook and tool session events)
packages/coding-agent/src/modes/interactive/interactive-mode.ts:1830: // Reset via session (emits hook and tool session events)
01:39:55 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "session.*hook\|emitSession" packages/coding-agent/src --include="*.ts" | head -20
01:40:00 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "type.*session\|SessionEvent" packages/coding-agent/src/core/hooks --include="*.ts" | head -20
01:40:04 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/hooks/types.ts:95-144
01:40:09 AM

Looking at the SessionEventBase, the session events don't include the model. They include session entries and file paths. So hooks shouldn't break when there's no model.

+

The only hooks that might care about model are:

+
    +
  • agent_start / agent_end - but these only fire when the agent actually runs (which requires a model)
    • +
  • tool_call / tool_result - same, only fire during agent execution
    • +
+

The session events (start, switch, clear, branch, shutdown) don't involve the model.

+

So I think we're fine. The session "start" hook would fire on startup even without a model, but it doesn't expect a model in its event data. And the agent/tool hooks only fire when actually prompting, which requires a model anyway.

+
01:40:57 AM

does the session header include the model? or do we set a model changed event on session start somewhere?

+
01:40:57 AM

Good question. Let me check:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "model.*changed\|setModel\|model_changed" packages/coding-agent/src --include="*.ts" | head -20
01:41:02 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "header\|Header" packages/coding-agent/src/modes/interactive --include="*.ts" | head -20
01:41:06 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/modes/interactive/interactive-mode.ts:200-279
01:41:10 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "updateStatus\|StatusBar\|status" packages/coding-agent/src/modes/interactive/interactive-mode.ts | head -30
01:41:16 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "footer\|Footer" packages/coding-agent/src/modes/interactive/interactive-mode.ts | head -20
01:41:20 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/modes/interactive/components/footer.ts:1-80
01:41:26 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/modes/interactive/components/footer.ts:81-180
01:41:29 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/modes/interactive/components/footer.ts:181-290
01:41:35 AM

I see the footer handles this.state.model?.id || "no-model" at line 232 and this.state.model?.contextWindow || 0 at line 167. So the footer already handles undefined model gracefully.

+

Let me also check the model changed event:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "modelFallbackMessage" packages/coding-agent/src --include="*.ts" | head -20
01:41:41 AM

The modelFallbackMessage is shown via mode.showWarning(). Let me check that the warning shows properly:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/main.ts:56-85
01:41:47 AM

Good, the modelFallbackMessage is displayed as a warning. So when there's no model, users will see "No models available. Use /login or set an API key environment variable."

+

The footer shows "no-model" gracefully and everything should work. I think we're good - the hooks don't expect a model in their event data, and the UI handles undefined model properly.

+
01:52:49 AM

alright, made some more minro chanbges. commit and push, comment with a concise "Thanks for reporting, fixed in the next release"

+
01:52:49 AM
$ cd /Users/badlogic/workspaces/pi-mono && git diff
01:52:54 AM
$ cd /Users/badlogic/workspaces/pi-mono && git add -A && git commit -m "Allow startup without API keys, fixes #288" && git push
01:53:04 AM
$ cd /Users/badlogic/workspaces/pi-mono && gh issue comment 288 --body "Thanks for reporting, fixed in the next release."
https://github.com/badlogic/pi-mono/issues/288#issuecomment-3688308754
01:53:11 AM

Done. Committed, pushed, and commented on the issue.

+
01:57:57 AM

https://github.com/badlogic/pi-mono/issues/295

+

any clean suggestion for this? what would it take in terms of changes to the code base? we do evn var resolution in the ai package iirc, I suppose we can make sdk.ts so it looks into the settings first and returns a getApiKey that first checkes settings for api keys, then delegates to what we do atm, based on provider?

+
01:57:57 AM
$ cd /Users/badlogic/workspaces/pi-mono && gh issue view 295 --comments
01:58:04 AM
$ cd /Users/badlogic/workspaces/pi-mono && gh issue view 295
01:58:10 AM

Let me look at how API keys are currently resolved:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "getApiKey\|ApiKey" packages/ai/src --include="*.ts" | head -30
01:58:15 AM
read ~/workspaces/pi-mono/packages/ai/src/stream.ts:19-98
01:58:18 AM

Now let me check the settings manager to see its structure:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "class SettingsManager\|interface.*Settings" packages/coding-agent/src/core/settings.ts | head -10
grep: packages/coding-agent/src/core/settings.ts: No such file or directory
01:58:22 AM
$ cd /Users/badlogic/workspaces/pi-mono && find packages/coding-agent/src -name "*.ts" | xargs grep -l "SettingsManager" | head -5
packages/coding-agent/src/core/model-resolver.ts
packages/coding-agent/src/core/settings-manager.ts
packages/coding-agent/src/core/sdk.ts
packages/coding-agent/src/core/agent-session.ts
packages/coding-agent/src/main.ts
01:58:26 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts:1-100
01:58:30 AM

Now let me check where defaultGetApiKey is used in sdk.ts:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "defaultGetApiKey\|getApiKey" packages/coding-agent/src/core/sdk.ts | head -20
01:58:36 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "getApiKeyForModel" packages/coding-agent/src/core/model-resolver.ts
9:import { findModel, getApiKeyForModel, getAvailableModels } from "./model-config.js";
314: const hasApiKey = restoredModel ? !!(await getApiKeyForModel(restoredModel)) : false;
01:58:40 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "getApiKeyForModel" packages/coding-agent/src/core/model-config.ts
274:export async function getApiKeyForModel(model: Model<Api>): Promise<string | undefined> {
373: const apiKey = await getApiKeyForModel(model);
01:58:43 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/model-config.ts:270-299
01:58:47 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/model-config.ts:300-349
01:58:51 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/model-config.ts:350-379
01:58:54 AM

Now I understand the structure. Here's my suggestion:

+

Clean approach:

+
    +
  1. Add an apiKeys field to Settings in settings-manager.ts:
    2. +
+
```typescript
export interface Settings {
+    // ... existing fields
+    apiKeys?: Record<string, string>;  // provider -> API key
+}
```
    +
  1. Modify getApiKeyForModel in model-config.ts to accept an optional settings-based key resolver, or create a wrapper in sdk.ts.
    2. +
+

The cleanest approach would be to:

+
    +
  1. Add apiKeys?: Record<string, string> to the Settings interface
    2. +
  2. Add getter/setter methods to SettingsManager for API keys
    3. +
  3. In sdk.ts, modify defaultGetApiKey() to first check settings, then fall back to the current behavior
    4. +
+

The change flow would be:

+
    +
  • sdk.ts creates a getApiKey function that wraps getApiKeyForModel
    • +
  • Before calling getApiKeyForModel, it checks settingsManager.getApiKey(model.provider)
    • +
  • If found in settings, return it; otherwise delegate to getApiKeyForModel
    • +
+

This keeps the ai package unchanged (it doesn't need to know about coding-agent settings) and puts the settings lookup in the coding-agent layer where it belongs.

+

Estimated changes:

+
    +
  • settings-manager.ts: ~15 lines (add interface field + getter/setter)
    • +
  • sdk.ts: ~10-15 lines (modify defaultGetApiKey or pass settings to the key resolver)
    • +
+

Want me to implement this?

+
02:00:12 AM

modify defaultGetApiKey

+
02:00:12 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts:30-59
02:00:16 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts
...
46 hookTimeout?: number; // Timeout for hook execution in ms (default: 30000)
47 customTools?: string[]; // Array of custom tool file paths
48 skills?: SkillsSettings;
49 terminal?: TerminalSettings;
+ 50 apiKeys?: Record<string, string>; // provider -> API key (e.g., { "anthropic": "sk-..." })
50 }
51
52 /** Deep merge settings: project/overrides take precedence, nested objects merge recursively */
53 function deepMergeSettings(base: Settings, overrides: Settings): Settings {
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts. Changed 750 characters to 842 characters.
02:00:26 AM

Now let me add a getter method to SettingsManager:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts:200-279
02:00:30 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts:280-379
02:00:37 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts
...
365 }
366 this.globalSettings.terminal.showImages = show;
367 this.save();
368 }
+369
+370 getApiKey(provider: string): string | undefined {
+371 return this.settings.apiKeys?.[provider];
+372 }
+373
+374 setApiKey(provider: string, key: string): void {
+375 if (!this.globalSettings.apiKeys) {
+376 this.globalSettings.apiKeys = {};
+377 }
+378 this.globalSettings.apiKeys[provider] = key;
+379 this.save();
+380 }
+381
+382 removeApiKey(provider: string): void {
+383 if (this.globalSettings.apiKeys) {
+384 delete this.globalSettings.apiKeys[provider];
+385 this.save();
+386 }
+387 }
+388
+389 getApiKeys(): Record<string, string> {
+390 return this.settings.apiKeys ?? {};
+391 }
369 }
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/settings-manager.ts. Changed 273 characters to 799 characters.
02:00:45 AM

Now let me modify defaultGetApiKey in sdk.ts to check settings first:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:325-354
02:00:50 AM

Now I see the issue. defaultGetApiKey doesn't have access to settingsManager. Let me check how it's used:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:540-559
02:00:54 AM

So settingsManager is already available in createAgentSession. I can pass it to defaultGetApiKey:

+
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
327 // API Key Helpers
328
329 /**
330 * Create the default API key resolver.
-331 * Checks custom providers (models.json), OAuth, and environment variables.
+331 * Priority: settings.json apiKeys > custom providers (models.json) > OAuth > environment variables.
332 */
-333 export function defaultGetApiKey(): (model: Model<any>) => Promise<string | undefined> {
-334 return getApiKeyForModel;
+333 export function defaultGetApiKey(
+334 settingsManager?: SettingsManager,
+335 ): (model: Model<any>) => Promise<string | undefined> {
+336 return async (model: Model<any>) => {
+337 // Check settings.json apiKeys first
+338 if (settingsManager) {
+339 const settingsKey = settingsManager.getApiKey(model.provider);
+340 if (settingsKey) {
+341 return settingsKey;
+342 }
+343 }
+344 // Fall back to existing resolution (custom providers, OAuth, env vars)
+345 return getApiKeyForModel(model);
+346 };
335 }
336
337 // System Prompt
338
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 261 characters to 633 characters.
02:01:03 AM

Now update the call site:

+
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
556 if (!model || !model.reasoning) {
557 thinkingLevel = "off";
558 }
559
-560 const getApiKey = options.getApiKey ?? defaultGetApiKey();
+560 const getApiKey = options.getApiKey ?? defaultGetApiKey(settingsManager);
561
562 const skills = options.skills ?? discoverSkills(cwd, agentDir, settingsManager.getSkillsSettings());
563 time("discoverSkills");
564
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 59 characters to 74 characters.
02:01:10 AM

Now I need to check if getApiKeyForModel is also used elsewhere for model discovery (checking if models are available). Let me check:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "getApiKeyForModel" packages/coding-agent/src/core/sdk.ts
45: getApiKeyForModel,
345: return getApiKeyForModel(model);
503: const key = await getApiKeyForModel(restoredModel);
520: const key = await getApiKeyForModel(settingsModel);
02:01:15 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:498-532
02:01:20 AM

These calls to getApiKeyForModel are for model discovery (checking if a model has a valid key). They should also check settings first. Let me create a helper that wraps this:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:470-504
02:01:26 AM

I should create a local helper function that checks settings first, then use it for model resolution:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:478-537
02:01:32 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
487 time("settingsManager");
488 const sessionManager = options.sessionManager ?? SessionManager.create(cwd, agentDir);
489 time("sessionManager");
490
+491 // Helper to check API key availability (settings first, then OAuth/env vars)
+492 const hasApiKey = async (m: Model<any>): Promise<boolean> => {
+493 const settingsKey = settingsManager.getApiKey(m.provider);
+494 if (settingsKey) return true;
+495 return !!(await getApiKeyForModel(m));
+496 };
+497
491 // Check if session has existing data to restore
492 const existingSession = sessionManager.loadSession();
493 time("loadSession");
494 const hasExistingSession = existingSession.messages.length > 0;
495
496 let model = options.model;
497 let modelFallbackMessage: string | undefined;
498
499 // If session has data, try to restore model from it
500 if (!model && hasExistingSession && existingSession.model) {
501 const restoredModel = findModel(existingSession.model.provider, existingSession.model.modelId);
-502 if (restoredModel) {
-503 const key = await getApiKeyForModel(restoredModel);
-504 if (key) {
-505 model = restoredModel;
-506 }
+509 if (restoredModel && (await hasApiKey(restoredModel))) {
+510 model = restoredModel;
507 }
508 if (!model) {
509 modelFallbackMessage = `Could not restore model ${existingSession.model.provider}/${existingSession.model.modelId}`;
510 }
511 }
512
513 // If still no model, try settings default
514 if (!model) {
515 const defaultProvider = settingsManager.getDefaultProvider();
516 const defaultModelId = settingsManager.getDefaultModel();
517 if (defaultProvider && defaultModelId) {
518 const settingsModel = findModel(defaultProvider, defaultModelId);
-519 if (settingsModel) {
-520 const key = await getApiKeyForModel(settingsModel);
-521 if (key) {
-522 model = settingsModel;
-523 }
+523 if (settingsModel && (await hasApiKey(settingsModel))) {
+524 model = settingsModel;
524 }
525 }
526 }
527
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 1757 characters to 1958 characters.
02:01:48 AM

Now I need to also make discoverAvailableModels aware of settings API keys. Let me check where that function is:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "discoverAvailableModels" packages/coding-agent/src/core/sdk.ts
225:export async function discoverAvailableModels(agentDir: string = getDefaultAgentDir()): Promise<Model<any>[]> {
531: const available = await discoverAvailableModels();
532: time("discoverAvailableModels");
02:01:53 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:220-249
02:01:56 AM

The issue is that discoverAvailableModels calls getAvailableModels from model-config.ts, which uses getApiKeyForModel directly. To make this work properly, I need to either:

+
    +
  1. Pass settingsManager through to discoverAvailableModels and down to getAvailableModels
    2. +
  2. Or use a different approach for checking available models in createAgentSession
    3. +
+

Let me check how getAvailableModels works:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/model-config.ts:360-399
02:02:03 AM

I think the cleanest approach here is to add an optional apiKeyResolver parameter to both discoverAvailableModels and getAvailableModels. But that's more invasive.

+

A simpler approach: since we already have hasApiKey helper in createAgentSession, let me just get all models and filter them there instead of using discoverAvailableModels:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "loadAndMergeModels\|getAllModels" packages/coding-agent/src/core/sdk.ts
47: loadAndMergeModels,
215: const { models, error } = loadAndMergeModels(agentDir);
02:02:08 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:210-234
02:02:12 AM

Let me update the code to use discoverModels and filter with hasApiKey:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:525-549
02:02:16 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
525 }
526 }
527 }
528
-529 // Fall back to first available
+529 // Fall back to first available model with a valid API key
530 if (!model) {
-531 const available = await discoverAvailableModels();
+531 const allModels = discoverModels(agentDir);
+532 for (const m of allModels) {
+533 if (await hasApiKey(m)) {
+534 model = m;
+535 break;
+536 }
+537 }
532 time("discoverAvailableModels");
-533 if (available.length > 0) {
-534 model = available[0];
+539 if (model) {
535 if (modelFallbackMessage) {
536 modelFallbackMessage += `. Using ${model.provider}/${model.id}`;
537 }
538 } else {
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 496 characters to 571 characters.
02:02:23 AM

Now let's run the check:

+
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
02:02:32 AM

Let me also export SettingsManager type from settings-manager.ts if needed, and check if the type is exported:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "export.*SettingsManager" packages/coding-agent/src/core/settings-manager.ts
84:export class SettingsManager {
02:02:38 AM

Good, it's exported. Now let me add the changelog entry:

+
read ~/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md:1-15
02:02:44 AM
edit ~/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md
1 # Changelog
2
3 ## [Unreleased]
4
+ 5 ### Added
+ 6
+ 7 - **API keys in settings.json**: Store API keys in `~/.pi/agent/settings.json` under the `apiKeys` field (e.g., `{ "apiKeys": { "anthropic": "sk-..." } }`). Settings keys take priority over environment variables. ([#295](https://github.com/badlogic/pi-mono/issues/295))
+ 8
5 ### Fixed
6
7 - **Allow startup without API keys**: Interactive mode no longer throws when no API keys are configured. Users can now start the agent and use `/login` to authenticate. ([#288](https://github.com/badlogic/pi-mono/issues/288))
8 - **`--system-prompt` file path support**: The `--system-prompt` argument now correctly resolves file paths (like `--append-system-prompt` already did). ([#287](https://github.com/badlogic/pi-mono/pull/287) by [@scutifer](https://github.com/scutifer))
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md. Changed 518 characters to 800 characters.
02:02:54 AM

Let me verify the changes look correct:

+
$ cd /Users/badlogic/workspaces/pi-mono && git diff --stat
packages/coding-agent/CHANGELOG.md | 4 ++
packages/coding-agent/src/core/sdk.ts | 54 ++++++++++++++--------
packages/coding-agent/src/core/settings-manager.ts | 24 ++++++++++
3 files changed, 64 insertions(+), 18 deletions(-)
02:02:59 AM
$ cd /Users/badlogic/workspaces/pi-mono && git diff packages/coding-agent/src/core/settings-manager.ts
02:03:02 AM
$ cd /Users/badlogic/workspaces/pi-mono && git diff packages/coding-agent/src/core/sdk.ts
02:03:05 AM

The changes look good. Let me run the check once more to make sure everything is fine:

+
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
02:03:14 AM

All checks pass. Here's a summary of the changes:

+

Summary:

+
    +

  1. settings-manager.ts: Added apiKeys?: Record<string, string> to Settings interface, plus getter/setter methods (getApiKey, setApiKey, removeApiKey, getApiKeys)

    +
    2. +

  2. sdk.ts:

    +
      +
    • Modified defaultGetApiKey() to accept optional SettingsManager and check settings API keys first before falling back to getApiKeyForModel
      • +
    • Added hasApiKey helper in createAgentSession that checks settings first
      • +
    • Updated model discovery to use hasApiKey helper (for session restore, settings default, and fallback)
      • +
    • Passed settingsManager to defaultGetApiKey()
      • +
    +
    3. +
+

Priority order for API keys:

+
    +
  1. settings.json apiKeys
    2. +
  2. Custom providers (models.json)
    3. +
  3. OAuth credentials
    4. +
  4. Environment variables
    5. +
+

Users can now add API keys to ~/.pi/agent/settings.json:

+
```json
{
+  "apiKeys": {
+    "anthropic": "sk-ant-...",
+    "openai": "sk-..."
+  }
+}
```
02:08:07 AM

well, it's unfortunate that we have to put settingsManager into defaultGetApiKey, but i see no better solution for now. we need to also update the api key seection in sdk.md

+
02:08:07 AM
read ~/workspaces/pi-mono/packages/coding-agent/docs/sdk.md:1-50
02:08:15 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "API Key\|api.*key\|getApiKey" packages/coding-agent/docs/sdk.md
259:### API Keys
269: getApiKey: async (model) => {
287:> See [examples/sdk/09-api-keys-and-oauth.ts](../examples/sdk/09-api-keys-and-oauth.ts)
714:const getApiKey = async (model: { provider: string }) => {
757: getApiKey,
02:08:21 AM
read ~/workspaces/pi-mono/packages/coding-agent/docs/sdk.md:255-299
02:08:27 AM
edit ~/workspaces/pi-mono/packages/coding-agent/docs/sdk.md
...
257 > See [examples/sdk/02-custom-model.ts](../examples/sdk/02-custom-model.ts)
258
259 ### API Keys
260
+261 API key resolution priority:
+262 1. `settings.json` apiKeys (e.g., `{ "apiKeys": { "anthropic": "sk-..." } }`)
+263 2. Custom providers from `models.json`
+264 3. OAuth credentials from `oauth.json`
+265 4. Environment variables (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, etc.)
+266
261 ```typescript
262 import { defaultGetApiKey, configureOAuthStorage } from "@mariozechner/pi-coding-agent";
263
-264 // Default: checks models.json, OAuth, environment variables
+270 // Default: checks settings.json, models.json, OAuth, environment variables
265 const { session } = await createAgentSession();
266
267 // Custom resolver
268 const { session } = await createAgentSession({
269 getApiKey: async (model) => {
270 // Custom logic (secrets manager, database, etc.)
271 if (model.provider === "anthropic") {
272 return process.env.MY_ANTHROPIC_KEY;
273 }
-274 // Fall back to default
+280 // Fall back to default (pass settingsManager for settings.json lookup)
275 return defaultGetApiKey()(model);
276 },
277 });
278
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/docs/sdk.md. Changed 912 characters to 1232 characters.
02:08:39 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
02:08:48 AM

Done. Updated the SDK docs to document the API key resolution priority.

+
02:11:13 AM

ok comimt and push close the related issue via "fixed #issue"

+
02:11:13 AM
$ cd /Users/badlogic/workspaces/pi-mono && git add -A && git commit -m "Add API keys in settings.json, fixes #295" && git push
02:11:23 AM

Done. Committed and pushed. The issue #295 will be automatically closed.

+
02:13:32 AM

i think README.md of coding-agent also needs an update?

+
02:13:32 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "API\|api.*key\|ANTHROPIC\|OPENAI" packages/coding-agent/README.md | head -30
02:13:36 AM
read ~/workspaces/pi-mono/packages/coding-agent/README.md:105-159
02:13:40 AM
edit ~/workspaces/pi-mono/packages/coding-agent/README.md
...
105 ```
106
107 ### API Keys & OAuth
108
-109 Set the environment variable for your provider:
+109 **Option 1: Settings file** (recommended)
110
+111 Add API keys to `~/.pi/agent/settings.json`:
+112
+113 ```json
+114 {
+115 "apiKeys": {
+116 "anthropic": "sk-ant-...",
+117 "openai": "sk-..."
+118 }
+119 }
+120 ```
+121
+122 **Option 2: Environment variables**
+123
111 | Provider | Environment Variable |
112 |----------|---------------------|
113 | Anthropic | `ANTHROPIC_API_KEY` |
114 | OpenAI | `OPENAI_API_KEY` |
115 | Google | `GEMINI_API_KEY` |
116 | Mistral | `MISTRAL_API_KEY` |
117 | Groq | `GROQ_API_KEY` |
118 | Cerebras | `CEREBRAS_API_KEY` |
119 | xAI | `XAI_API_KEY` |
120 | OpenRouter | `OPENROUTER_API_KEY` |
121 | ZAI | `ZAI_API_KEY` |
122
+136 Settings file keys take priority over environment variables.
+137
123 **OAuth Providers:**
124
125 Use `/login` to authenticate with subscription-based or free-tier providers:
126
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/README.md. Changed 415 characters to 644 characters.
02:13:49 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
02:13:57 AM
$ cd /Users/badlogic/workspaces/pi-mono && git add -A && git commit -m "Update README with settings.json API keys" && git push
02:14:07 AM

Done.

+
02:21:59 AM

well, getApiKeyForModel is called in various places, and that one does not take into account the apikeys in settings. we only set the apikey function on the transport, but lots of places in the coding-agent actually don't use the api key functino from the transport, but instead the global getApiKeyForModel, which only checks cutom model keys, oauth, and the ai package env vars. i guess the issue is that we store that shit globally ...

+
02:21:59 AM

You're right. The problem is that getApiKeyForModel in model-config.ts is used directly in many places (model discovery, model switching, etc.) and it doesn't know about settings.

+

Looking at the options:

+
    +

  1. Global state in model-config.ts - Add a setSettingsApiKeyResolver that gets called once during init, similar to how configureOAuthStorage works. Then getApiKeyForModel checks that first.

    +
    2. +

  2. Pass settingsManager everywhere - Invasive, touches many files.

    +
    3. +

  3. Merge apiKeys into the existing apiKeys Map in @mariozechner/pi-ai - Call setApiKey() from the ai package for each key in settings during startup. Simple but slightly weird since settings can change.

    +
    4. +
+

Option 1 seems cleanest. Something like:

+
```typescript
// In model-config.ts
+let settingsApiKeyResolver: ((provider: string) => string | undefined) | null = null;
+
+export function setSettingsApiKeyResolver(resolver: (provider: string) => string | undefined): void {
+  settingsApiKeyResolver = resolver;
+}
+
+export async function getApiKeyForModel(model: Model<Api>): Promise<string | undefined> {
+  // Check settings first
+  if (settingsApiKeyResolver) {
+    const key = settingsApiKeyResolver(model.provider);
+    if (key) return key;
+  }
+  // ... rest of existing logic
+}
```

Then in createAgentSession:

+
```typescript
setSettingsApiKeyResolver((provider) => settingsManager.getApiKey(provider));
```

This mirrors the pattern used for OAuth storage configuration. Want me to implement this?

+
02:23:12 AM

if we createAgentSession multiple times in the same process, with different getApiKey option, then this fucks up. the global state is not good :/

+
02:23:12 AM

Yeah, you're right. The global state is fundamentally problematic for multi-session scenarios.

+

Options:

+
    +

  1. Accept the limitation - Document that apiKeys in settings.json only works for the CLI, not for SDK users who create multiple sessions with different key resolvers. SDK users should use the getApiKey option explicitly.

    +
    2. +

  2. Thread settingsManager through everything - Invasive refactor, getApiKeyForModel becomes getApiKeyForModel(model, settingsManager?), ripples through model-config, model-resolver, etc.

    +
    3. +

  3. Make settings lookup part of the ai package - Add a setApiKeyResolver callback there that gets checked before env vars. Still global, same problem.

    +
    4. +

  4. Lazy approach - In createAgentSession, iterate settings apiKeys and call setApiKey(provider, key) from the ai package. This populates the global apiKeys Map. Problem: last session wins if they have different settings.

    +
    5. +
+

Realistically, for the CLI use case (single session per process), option 1 is probably fine. The current implementation already works for the transport's getApiKey callback, which is what matters for actual API calls. The issue is just model discovery/availability checks.

+

For model discovery, we could document that settings.json apiKeys are used for the active session but model discovery at startup uses env vars/OAuth. Or we accept the slight inconsistency.

+

What's your preference?

+
02:25:04 AM

well, the ai package handles oauth, righ? how?

+
02:25:04 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -rn "oauth\|OAuth" packages/ai/src --include="*.ts" | grep -v node_modules | head -30
02:25:10 AM
read ~/workspaces/pi-mono/packages/ai/src/utils/oauth/storage.ts:1-80
02:25:16 AM
read ~/workspaces/pi-mono/packages/ai/src/utils/oauth/storage.ts:75-154
02:25:21 AM

I see. OAuth uses setOAuthStorage() to configure a global backend. Same global state pattern. The difference is OAuth storage is file-based and doesn't change per-session typically.

+

For API keys, we could do the same: add a setApiKeyResolver in the ai package that getApiKey/resolveApiKey checks first. It's still global, but:

+
    +
  1. For CLI (single session per process), it works fine
    2. +
  2. For SDK with multiple sessions, the last one wins (same as OAuth currently)
    3. +
+

The alternative is accepting that settings.json apiKeys only affect the transport's getApiKey callback (actual API calls), not model discovery. Model discovery would still use env vars/OAuth. This is a minor inconsistency but probably acceptable in practice.

+

What do you think? Add a global setApiKeyResolver to the ai package, matching the OAuth pattern?

+
02:27:15 AM

why does the ai package need that global state? don't we just pass in a validated/refreshed api key to streamSimple/complete/whatever? oauth should be outside of all that machinery. also, where is model discovery handled? why is there model-config.ts and model-resolver.ts?

+
02:27:15 AM

You're right to question this. Let me check:

+
$ cd /Users/badlogic/workspaces/pi-mono && head -50 packages/ai/src/stream.ts
02:27:20 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "getApiKey\|resolveApiKey" packages/ai/src/stream.ts
02:27:25 AM
$ cd /Users/badlogic/workspaces/pi-mono && head -30 packages/coding-agent/src/core/model-config.ts
02:27:31 AM
$ cd /Users/badlogic/workspaces/pi-mono && head -30 packages/coding-agent/src/core/model-resolver.ts
02:27:34 AM

So model-config.ts handles:

+
    +
  • Loading/merging models from built-in + models.json
    • +
  • getApiKeyForModel - resolving API keys (custom providers, OAuth, env vars)
    • +
  • Model availability checking
    • +
+

And model-resolver.ts handles:

+
    +
  • CLI argument parsing for models (--model provider/id:thinking)
    • +
  • Model scoping/selection logic
    • +
  • Default model per provider
    • +
+

The ai package's getApiKey/resolveApiKey is used by stream()/streamSimple() when no apiKey option is passed. But in coding-agent, we always pass the apiKey via the transport's getApiKey callback.

+

So the ai package's global getApiKey is really only used:

+
    +
  1. By model-config.ts's getApiKeyForModel as a fallback for built-in providers
    2. +
  2. Directly by stream()/streamSimple() when called without explicit apiKey (not our case)
    3. +
+

The real issue is getApiKeyForModel in model-config.ts is called from many places for model discovery/availability. We could refactor it to accept an optional resolver, or we accept that settings.json keys only affect the transport callback.

+

Want me to refactor getApiKeyForModel to accept an optional settingsApiKeyResolver parameter?

+
02:28:48 AM

agent-session.ts also calls getApiKeyForModel

+
02:28:48 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "getApiKeyForModel" packages/coding-agent/src/core/agent-session.ts
02:28:52 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts:555-574
02:28:55 AM

I see. AgentSession has access to settingsManager (it's passed in during construction). So we could:

+
    +
  1. Add a method on AgentSession that wraps getApiKeyForModel and checks settings first
    2. +
  2. Or refactor getApiKeyForModel to accept an optional resolver
    3. +
+

Actually, AgentSession already has settingsManager. The cleanest fix would be to add a private helper in AgentSession that checks settings first:

+
```typescript
private async resolveApiKey(model: Model<any>): Promise<string | undefined> {
+    const settingsKey = this.settingsManager.getApiKey(model.provider);
+    if (settingsKey) return settingsKey;
+    return getApiKeyForModel(model);
+}
```

Then replace all getApiKeyForModel calls in agent-session.ts with this.resolveApiKey.

+

For model-config.ts and sdk.ts, they already have access to settingsManager in the relevant contexts.

+

This keeps the change localized and doesn't require global state. Want me to do this?

+
02:33:40 AM

sounds good, do it

+
02:33:40 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts:140-179
02:33:46 AM
read ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts:160-189
02:33:51 AM
Aborted
02:34:31 AM

actually, i think AgentSessionConfig should take a resolveApiKey(provider: string)?

+
02:34:31 AM

Good idea. That's cleaner - the session gets its resolver injected, and createAgentSession can construct it with settings awareness.

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts:40-69
02:34:36 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts
...
55 hookRunner?: HookRunner | null;
56 /** Custom tools for session lifecycle events */
57 customTools?: LoadedCustomTool[];
58 skillsSettings?: Required<SkillsSettings>;
+ 59 /** Resolve API key for a model. Default: getApiKeyForModel */
+ 60 resolveApiKey?: (model: Model<any>) => Promise<string | undefined>;
59 }
60
61 /** Options for AgentSession.prompt() */
62 export interface PromptOptions {
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts. Changed 569 characters to 702 characters.
02:34:46 AM

Now add the field and use it in the constructor:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts:130-169
02:34:51 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts
...
152 private _customTools: LoadedCustomTool[] = [];
153
154 private _skillsSettings: Required<SkillsSettings> | undefined;
155
+ 156 // API key resolver
+ 157 private _resolveApiKey: (model: Model<any>) => Promise<string | undefined>;
+ 158
156 constructor(config: AgentSessionConfig) {
157 this.agent = config.agent;
158 this.sessionManager = config.sessionManager;
159 this.settingsManager = config.settingsManager;
160 this._scopedModels = config.scopedModels ?? [];
161 this._fileCommands = config.fileCommands ?? [];
162 this._hookRunner = config.hookRunner ?? null;
163 this._customTools = config.customTools ?? [];
164 this._skillsSettings = config.skillsSettings;
+ 168 this._resolveApiKey = config.resolveApiKey ?? getApiKeyForModel;
165 }
166
167 // =========================================================================
168 // Event Subscription
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/agent-session.ts. Changed 567 characters to 733 characters.
02:34:59 AM

Now replace all calls to getApiKeyForModel with this._resolveApiKey:

+
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "await getApiKeyForModel" packages/coding-agent/src/core/agent-session.ts
02:35:03 AM
$ cd /Users/badlogic/workspaces/pi-mono && sed -i '' 's/await getApiKeyForModel(/await this._resolveApiKey(/g' packages/coding-agent/src/core/agent-session.ts
(no output)
02:35:07 AM
$ cd /Users/badlogic/workspaces/pi-mono && grep -n "getApiKeyForModel" packages/coding-agent/src/core/agent-session.ts
26:import { getApiKeyForModel, getAvailableModels } from "./model-config.js";
59: /** Resolve API key for a model. Default: getApiKeyForModel */
168: this._resolveApiKey = config.resolveApiKey ?? getApiKeyForModel;
02:35:10 AM

Good, only the import and default remain. Now update sdk.ts to pass the resolver:

+
read ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts:665-694
02:35:15 AM
edit ~/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts
...
680 fileCommands: slashCommands,
681 hookRunner,
682 customTools: customToolsResult.tools,
683 skillsSettings: settingsManager.getSkillsSettings(),
+684 resolveApiKey: getApiKey,
684 });
685 time("createAgentSession");
686
687 return {
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/src/core/sdk.ts. Changed 293 characters to 321 characters.
02:35:21 AM

Now let's run the check:

+
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
02:35:29 AM

All checks pass. Let me verify the changes look correct:

+
$ cd /Users/badlogic/workspaces/pi-mono && git diff --stat
packages/coding-agent/src/core/agent-session.ts | 18 ++++++++++++------
packages/coding-agent/src/core/sdk.ts | 1 +
2 files changed, 13 insertions(+), 6 deletions(-)
02:35:32 AM
$ cd /Users/badlogic/workspaces/pi-mono && git diff
02:35:36 AM

Looks good. Now AgentSession uses the injected resolveApiKey which in createAgentSession is set to the getApiKey function that checks settings first.

+
$ cd /Users/badlogic/workspaces/pi-mono && git add -A && git commit -m "Use injected resolveApiKey in AgentSession for settings.json support" && git push
02:35:47 AM

Done. Now AgentSession gets its API key resolver injected, and createAgentSession passes one that checks settings.json first. This avoids the global state problem.

+
02:42:13 AM
```json
{
+  "apiKeys": {
+    "anthropic": "sk-ant-...",
+    "openai": "sk-..."
+  }
+}
+
+should list all the key names here in README.md
```
02:42:13 AM
read ~/workspaces/pi-mono/packages/coding-agent/README.md:105-144
02:42:20 AM
edit ~/workspaces/pi-mono/packages/coding-agent/README.md
...
113 ```json
114 {
115 "apiKeys": {
116 "anthropic": "sk-ant-...",
-117 "openai": "sk-..."
+117 "openai": "sk-...",
+118 "google": "...",
+119 "mistral": "...",
+120 "groq": "...",
+121 "cerebras": "...",
+122 "xai": "...",
+123 "openrouter": "...",
+124 "zai": "..."
118 }
119 }
120 ```
121
122 **Option 2: Environment variables**
123
-124 | Provider | Environment Variable |
-125 |----------|---------------------|
-126 | Anthropic | `ANTHROPIC_API_KEY` |
-127 | OpenAI | `OPENAI_API_KEY` |
-128 | Google | `GEMINI_API_KEY` |
-129 | Mistral | `MISTRAL_API_KEY` |
-130 | Groq | `GROQ_API_KEY` |
-131 | Cerebras | `CEREBRAS_API_KEY` |
-132 | xAI | `XAI_API_KEY` |
-133 | OpenRouter | `OPENROUTER_API_KEY` |
-134 | ZAI | `ZAI_API_KEY` |
+131 | Provider | Settings Key | Environment Variable |
+132 |----------|--------------|---------------------|
+133 | Anthropic | `anthropic` | `ANTHROPIC_API_KEY` |
+134 | OpenAI | `openai` | `OPENAI_API_KEY` |
+135 | Google | `google` | `GEMINI_API_KEY` |
+136 | Mistral | `mistral` | `MISTRAL_API_KEY` |
+137 | Groq | `groq` | `GROQ_API_KEY` |
+138 | Cerebras | `cerebras` | `CEREBRAS_API_KEY` |
+139 | xAI | `xai` | `XAI_API_KEY` |
+140 | OpenRouter | `openrouter` | `OPENROUTER_API_KEY` |
+141 | ZAI | `zai` | `ZAI_API_KEY` |
135
136 Settings file keys take priority over environment variables.
137
138 **OAuth Providers:**
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/README.md. Changed 622 characters to 899 characters.
02:42:30 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run check 2>&1
02:42:37 AM
$ cd /Users/badlogic/workspaces/pi-mono && git add -A && git commit -m "List all provider keys in README settings example" && git push
02:42:47 AM

Done.

+
02:43:41 AM

alright, open a new issue "Untangle OAuth/API key mess" with label coding-agent, and describe the curernt mess, what problems it has (global state), and a possible concise, simple refactor

+
02:43:41 AM
$ cd /Users/badlogic/workspaces/pi-mono && gh issue create --title "Untangle OAuth/API key mess" --label "pkg:coding-agent" --body '## Current State + +API key resolution is scattered across multiple layers with global state: + +1. **`@mariozechner/pi-ai` (`stream.ts`)**: Global `apiKeys` Map + `setApiKey()`/`getApiKey()` + `resolveApiKey()` that checks OAuth +2. **`@mariozechner/pi-ai` (`oauth/storage.ts`)**: Global `currentBackend` for OAuth storage, configured via `setOAuthStorage()` +3. **`coding-agent` (`model-config.ts`)**: `getApiKeyForModel()` that checks custom providers, OAuth, then delegates to ai package +4. **`coding-agent` (`sdk.ts`)**: `defaultGetApiKey()` that checks settings.json first, then `getApiKeyForModel()` +5. **`coding-agent` (`agent-session.ts`)**: Injected `resolveApiKey` to use settings-aware resolver + +## Problems + +1. **Global state breaks multi-session**: If you call `createAgentSession()` multiple times with different `getApiKey` options, the global state in ai package is shared +2. **Inconsistent resolution**: Model discovery in `model-config.ts` uses `getApiKeyForModel()` directly, which does not check settings.json apiKeys (fixed partially by using `hasApiKey` helper in `sdk.ts`) +3. **OAuth storage is global**: `setOAuthStorage()` affects all sessions in the process +4. **Confusing layering**: Hard to understand which function checks what and in which order + +## Proposed Refactor + +1. **Remove global state from ai package**: `stream()`/`streamSimple()` should require explicit `apiKey` parameter, no fallback to global map or env vars +2. **Move all key resolution to coding-agent**: Create a single `ApiKeyResolver` class/interface that handles: + - Settings.json apiKeys + - Custom providers from models.json + - OAuth credentials + - Environment variables +3. **Pass resolver explicitly**: `createAgentSession()` creates the resolver, passes it to `AgentSession`, and uses it for model discovery +4. **OAuth storage per-session**: Pass storage backend to resolver instead of global `setOAuthStorage()` + +This makes the ai package a pure streaming library with no global state, and consolidates all key resolution logic in coding-agent where it belongs.'
https://github.com/badlogic/pi-mono/issues/296
02:44:01 AM

Done. Created issue #296 describing the current mess and proposed refactor.

+
02:45:13 AM

ok, do a release

+
02:45:13 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run version:patch
02:46:06 AM
read ~/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md:1-20
02:46:09 AM
edit ~/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md
1 # Changelog
2
3 ## [Unreleased]
4
+ 5 ## [0.27.3] - 2025-12-24
+ 6
5 ### Added
6
7 - **API keys in settings.json**: Store API keys in `~/.pi/agent/settings.json` under the `apiKeys` field (e.g., `{ "apiKeys": { "anthropic": "sk-..." } }`). Settings keys take priority over environment variables. ([#295](https://github.com/badlogic/pi-mono/issues/295))
8
...
Successfully replaced text in /Users/badlogic/workspaces/pi-mono/packages/coding-agent/CHANGELOG.md. Changed 800 characters to 826 characters.
02:46:18 AM
$ cd /Users/badlogic/workspaces/pi-mono && git add -A && git commit -m "Release v0.27.3" && git tag v0.27.3 && git push origin main && git push origin v0.27.3
02:46:31 AM
$ cd /Users/badlogic/workspaces/pi-mono && npm run publish 2>&1
02:47:08 AM

Released v0.27.3 to npm.

+
+
+ +
+ Generated by pi coding-agent on 12/24/2025, 12:06:50 PM +
+
+ + \ No newline at end of file diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts index 048c43ac..62584252 100644 --- a/packages/ai/src/models.generated.ts +++ b/packages/ai/src/models.generated.ts @@ -6359,23 +6359,6 @@ export const MODELS = { contextWindow: 128000, maxTokens: 16384, } satisfies Model<"openai-completions">, - "meta-llama/llama-3.1-70b-instruct": { - id: "meta-llama/llama-3.1-70b-instruct", - name: "Meta: Llama 3.1 70B Instruct", - api: "openai-completions", - provider: "openrouter", - baseUrl: "https://openrouter.ai/api/v1", - reasoning: false, - input: ["text"], - cost: { - input: 0.39999999999999997, - output: 0.39999999999999997, - cacheRead: 0, - cacheWrite: 0, - }, - contextWindow: 131072, - maxTokens: 4096, - } satisfies Model<"openai-completions">, "meta-llama/llama-3.1-8b-instruct": { id: "meta-llama/llama-3.1-8b-instruct", name: "Meta: Llama 3.1 8B Instruct", @@ -6410,6 +6393,23 @@ export const MODELS = { contextWindow: 10000, maxTokens: 4096, } satisfies Model<"openai-completions">, + "meta-llama/llama-3.1-70b-instruct": { + id: "meta-llama/llama-3.1-70b-instruct", + name: "Meta: Llama 3.1 70B Instruct", + api: "openai-completions", + provider: "openrouter", + baseUrl: "https://openrouter.ai/api/v1", + reasoning: false, + input: ["text"], + cost: { + input: 0.39999999999999997, + output: 0.39999999999999997, + cacheRead: 0, + cacheWrite: 0, + }, + contextWindow: 131072, + maxTokens: 4096, + } satisfies Model<"openai-completions">, "mistralai/mistral-nemo": { id: "mistralai/mistral-nemo", name: "Mistral: Mistral Nemo", @@ -6546,6 +6546,23 @@ export const MODELS = { contextWindow: 128000, maxTokens: 4096, } satisfies Model<"openai-completions">, + "openai/gpt-4o-2024-05-13": { + id: "openai/gpt-4o-2024-05-13", + name: "OpenAI: GPT-4o (2024-05-13)", + api: "openai-completions", + provider: "openrouter", + baseUrl: "https://openrouter.ai/api/v1", + reasoning: false, + input: ["text", "image"], + cost: { + input: 5, + output: 15, + cacheRead: 0, + cacheWrite: 0, + }, + contextWindow: 128000, + maxTokens: 4096, + } satisfies Model<"openai-completions">, "openai/gpt-4o": { id: "openai/gpt-4o", name: "OpenAI: GPT-4o", @@ -6580,23 +6597,6 @@ export const MODELS = { contextWindow: 128000, maxTokens: 64000, } satisfies Model<"openai-completions">, - "openai/gpt-4o-2024-05-13": { - id: "openai/gpt-4o-2024-05-13", - name: "OpenAI: GPT-4o (2024-05-13)", - api: "openai-completions", - provider: "openrouter", - baseUrl: "https://openrouter.ai/api/v1", - reasoning: false, - input: ["text", "image"], - cost: { - input: 5, - output: 15, - cacheRead: 0, - cacheWrite: 0, - }, - contextWindow: 128000, - maxTokens: 4096, - } satisfies Model<"openai-completions">, "meta-llama/llama-3-70b-instruct": { id: "meta-llama/llama-3-70b-instruct", name: "Meta: Llama 3 70B Instruct", @@ -6716,23 +6716,6 @@ export const MODELS = { contextWindow: 128000, maxTokens: 4096, } satisfies Model<"openai-completions">, - "openai/gpt-4-turbo-preview": { - id: "openai/gpt-4-turbo-preview", - name: "OpenAI: GPT-4 Turbo Preview", - api: "openai-completions", - provider: "openrouter", - baseUrl: "https://openrouter.ai/api/v1", - reasoning: false, - input: ["text"], - cost: { - input: 10, - output: 30, - cacheRead: 0, - cacheWrite: 0, - }, - contextWindow: 128000, - maxTokens: 4096, - } satisfies Model<"openai-completions">, "openai/gpt-3.5-turbo-0613": { id: "openai/gpt-3.5-turbo-0613", name: "OpenAI: GPT-3.5 Turbo (older v0613)", @@ -6750,6 +6733,23 @@ export const MODELS = { contextWindow: 4095, maxTokens: 4096, } satisfies Model<"openai-completions">, + "openai/gpt-4-turbo-preview": { + id: "openai/gpt-4-turbo-preview", + name: "OpenAI: GPT-4 Turbo Preview", + api: "openai-completions", + provider: "openrouter", + baseUrl: "https://openrouter.ai/api/v1", + reasoning: false, + input: ["text"], + cost: { + input: 10, + output: 30, + cacheRead: 0, + cacheWrite: 0, + }, + contextWindow: 128000, + maxTokens: 4096, + } satisfies Model<"openai-completions">, "mistralai/mistral-tiny": { id: "mistralai/mistral-tiny", name: "Mistral Tiny", diff --git a/packages/coding-agent/src/core/export-html.ts b/packages/coding-agent/src/core/export-html.ts index ea48edc2..bff53487 100644 --- a/packages/coding-agent/src/core/export-html.ts +++ b/packages/coding-agent/src/core/export-html.ts @@ -1,6 +1,7 @@ import type { AgentState } from "@mariozechner/pi-agent-core"; import type { AssistantMessage, ImageContent, Message, ToolResultMessage, UserMessage } from "@mariozechner/pi-ai"; import { existsSync, readFileSync, writeFileSync } from "fs"; +import hljs from "highlight.js"; import { marked } from "marked"; import { homedir } from "os"; import * as path from "path"; @@ -322,17 +323,123 @@ function formatTimestamp(timestamp: number | string | undefined): string { return date.toLocaleTimeString(undefined, { hour: "2-digit", minute: "2-digit", second: "2-digit" }); } -/** Render markdown to HTML server-side with TUI-style code block formatting. */ +/** Highlight code using highlight.js. Returns HTML with syntax highlighting spans. */ +function highlightCode(code: string, lang?: string): string { + if (!lang) { + return escapeHtml(code); + } + try { + // Check if language is supported + if (hljs.getLanguage(lang)) { + return hljs.highlight(code, { language: lang, ignoreIllegals: true }).value; + } + // Try common aliases + const aliases: Record = { + ts: "typescript", + js: "javascript", + py: "python", + rb: "ruby", + sh: "bash", + yml: "yaml", + md: "markdown", + }; + const aliasedLang = aliases[lang]; + if (aliasedLang && hljs.getLanguage(aliasedLang)) { + return hljs.highlight(code, { language: aliasedLang, ignoreIllegals: true }).value; + } + } catch { + // Fall through to escaped output + } + return escapeHtml(code); +} + +/** Get language from file path extension. */ +function getLanguageFromPath(filePath: string): string | undefined { + const ext = filePath.split(".").pop()?.toLowerCase(); + if (!ext) return undefined; + + const extToLang: Record = { + ts: "typescript", + tsx: "typescript", + js: "javascript", + jsx: "javascript", + mjs: "javascript", + cjs: "javascript", + py: "python", + rb: "ruby", + rs: "rust", + go: "go", + java: "java", + kt: "kotlin", + swift: "swift", + c: "c", + h: "c", + cpp: "cpp", + cc: "cpp", + cxx: "cpp", + hpp: "cpp", + cs: "csharp", + php: "php", + sh: "bash", + bash: "bash", + zsh: "bash", + fish: "bash", + ps1: "powershell", + sql: "sql", + html: "html", + htm: "html", + xml: "xml", + css: "css", + scss: "scss", + sass: "scss", + less: "less", + json: "json", + yaml: "yaml", + yml: "yaml", + toml: "toml", + ini: "ini", + md: "markdown", + markdown: "markdown", + dockerfile: "dockerfile", + makefile: "makefile", + cmake: "cmake", + lua: "lua", + r: "r", + scala: "scala", + clj: "clojure", + cljs: "clojure", + ex: "elixir", + exs: "elixir", + erl: "erlang", + hrl: "erlang", + hs: "haskell", + ml: "ocaml", + mli: "ocaml", + fs: "fsharp", + fsx: "fsharp", + vue: "vue", + svelte: "xml", + tf: "hcl", + hcl: "hcl", + proto: "protobuf", + graphql: "graphql", + gql: "graphql", + }; + + return extToLang[ext]; +} + +/** Render markdown to HTML server-side with TUI-style code block formatting and syntax highlighting. */ function renderMarkdown(text: string): string { // Custom renderer for code blocks to match TUI style const renderer = new marked.Renderer(); renderer.code = ({ text: code, lang }: { text: string; lang?: string }) => { const language = lang || ""; - const escaped = escapeHtml(code); + const highlighted = highlightCode(code, lang); return ( '
' + `
\`\`\`${language}
` + - `
${escaped}
` + + `
${highlighted}
` + '
```
' + "
" ); @@ -348,10 +455,32 @@ function renderMarkdown(text: string): string { return marked.parse(text, { renderer }) as string; } -function formatExpandableOutput(lines: string[], maxLines: number): string { +function formatExpandableOutput(lines: string[], maxLines: number, lang?: string): string { const displayLines = lines.slice(0, maxLines); const remaining = lines.length - maxLines; + // If language is provided, highlight the entire code block + if (lang) { + const code = lines.join("\n"); + const highlighted = highlightCode(code, lang); + + if (remaining > 0) { + // For expandable, we need preview and full versions + const previewCode = displayLines.join("\n"); + const previewHighlighted = highlightCode(previewCode, lang); + + let out = '`; + return out; + } + + return `
${highlighted}
`; + } + + // No language - plain text output if (remaining > 0) { let out = '