mirror of
https://github.com/getcompanion-ai/co-mono.git
synced 2026-04-17 08:00:59 +00:00
Simplify compaction: remove proactive abort, use Agent.continue() for retry
- Add agentLoopContinue() to pi-ai for resuming from existing context - Add Agent.continue() method and transport.continue() interface - Simplify AgentSession compaction to two cases: overflow (auto-retry) and threshold (no retry) - Remove proactive mid-turn compaction abort - Merge turn prefix summary into main summary - Add isCompacting property to AgentSession and RPC state - Block input during compaction in interactive mode - Show compaction count on session resume - Rename RPC.md to rpc.md for consistency Related to #128
This commit is contained in:
parent
d67c69c6e9
commit
5a9d844f9a
27 changed files with 1261 additions and 1011 deletions
|
|
@ -11,7 +11,7 @@ import type {
|
|||
ToolCall,
|
||||
UserMessage,
|
||||
} from "@mariozechner/pi-ai";
|
||||
import { agentLoop } from "@mariozechner/pi-ai";
|
||||
import { agentLoop, agentLoopContinue } from "@mariozechner/pi-ai";
|
||||
import { AssistantMessageEventStream } from "@mariozechner/pi-ai/dist/utils/event-stream.js";
|
||||
import { parseStreamingJson } from "@mariozechner/pi-ai/dist/utils/json-parse.js";
|
||||
import { clearAuthToken, getAuthToken } from "../../utils/auth-token.js";
|
||||
|
|
@ -315,51 +315,57 @@ function streamSimpleProxy(
|
|||
return stream;
|
||||
}
|
||||
|
||||
// Proxy transport executes the turn using a remote proxy server
|
||||
/**
|
||||
* Transport that uses an app server with user authentication tokens.
|
||||
* The server manages user accounts and proxies requests to LLM providers.
|
||||
*/
|
||||
export class AppTransport implements AgentTransport {
|
||||
// Hardcoded proxy URL for now - will be made configurable later
|
||||
private readonly proxyUrl = "https://genai.mariozechner.at";
|
||||
|
||||
async *run(messages: Message[], userMessage: Message, cfg: AgentRunConfig, signal?: AbortSignal) {
|
||||
private async getStreamFn() {
|
||||
const authToken = await getAuthToken();
|
||||
if (!authToken) {
|
||||
throw new Error(i18n("Auth token is required for proxy transport"));
|
||||
}
|
||||
|
||||
// Use proxy - no local API key needed
|
||||
const streamFn = <TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions) => {
|
||||
return streamSimpleProxy(
|
||||
model,
|
||||
context,
|
||||
{
|
||||
...options,
|
||||
authToken,
|
||||
},
|
||||
this.proxyUrl,
|
||||
);
|
||||
return <TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions) => {
|
||||
return streamSimpleProxy(model, context, { ...options, authToken }, this.proxyUrl);
|
||||
};
|
||||
}
|
||||
|
||||
// Messages are already LLM-compatible (filtered by Agent)
|
||||
const context: AgentContext = {
|
||||
private buildContext(messages: Message[], cfg: AgentRunConfig): AgentContext {
|
||||
return {
|
||||
systemPrompt: cfg.systemPrompt,
|
||||
messages,
|
||||
tools: cfg.tools,
|
||||
};
|
||||
}
|
||||
|
||||
const pc: AgentLoopConfig = {
|
||||
private buildLoopConfig(cfg: AgentRunConfig): AgentLoopConfig {
|
||||
return {
|
||||
model: cfg.model,
|
||||
reasoning: cfg.reasoning,
|
||||
getQueuedMessages: cfg.getQueuedMessages,
|
||||
};
|
||||
}
|
||||
|
||||
async *run(messages: Message[], userMessage: Message, cfg: AgentRunConfig, signal?: AbortSignal) {
|
||||
const streamFn = await this.getStreamFn();
|
||||
const context = this.buildContext(messages, cfg);
|
||||
const pc = this.buildLoopConfig(cfg);
|
||||
|
||||
// Yield events from the upstream agentLoop iterator
|
||||
// Pass streamFn as the 5th parameter to use proxy
|
||||
for await (const ev of agentLoop(userMessage as unknown as UserMessage, context, pc, signal, streamFn as any)) {
|
||||
yield ev;
|
||||
}
|
||||
}
|
||||
|
||||
async *continue(messages: Message[], cfg: AgentRunConfig, signal?: AbortSignal) {
|
||||
const streamFn = await this.getStreamFn();
|
||||
const context = this.buildContext(messages, cfg);
|
||||
const pc = this.buildLoopConfig(cfg);
|
||||
|
||||
for await (const ev of agentLoopContinue(context, pc, signal, streamFn as any)) {
|
||||
yield ev;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -2,6 +2,7 @@ import {
|
|||
type AgentContext,
|
||||
type AgentLoopConfig,
|
||||
agentLoop,
|
||||
agentLoopContinue,
|
||||
type Message,
|
||||
type UserMessage,
|
||||
} from "@mariozechner/pi-ai";
|
||||
|
|
@ -14,37 +15,53 @@ import type { AgentRunConfig, AgentTransport } from "./types.js";
|
|||
* Uses CORS proxy only for providers that require it (Anthropic OAuth, Z-AI).
|
||||
*/
|
||||
export class ProviderTransport implements AgentTransport {
|
||||
async *run(messages: Message[], userMessage: Message, cfg: AgentRunConfig, signal?: AbortSignal) {
|
||||
// Get API key from storage
|
||||
private async getModelAndKey(cfg: AgentRunConfig) {
|
||||
const apiKey = await getAppStorage().providerKeys.get(cfg.model.provider);
|
||||
if (!apiKey) {
|
||||
throw new Error("no-api-key");
|
||||
}
|
||||
|
||||
// Get proxy URL from settings (if available)
|
||||
const proxyEnabled = await getAppStorage().settings.get<boolean>("proxy.enabled");
|
||||
const proxyUrl = await getAppStorage().settings.get<string>("proxy.url");
|
||||
|
||||
// Apply proxy only if this provider/key combination requires it
|
||||
const model = applyProxyIfNeeded(cfg.model, apiKey, proxyEnabled ? proxyUrl || undefined : undefined);
|
||||
|
||||
// Messages are already LLM-compatible (filtered by Agent)
|
||||
const context: AgentContext = {
|
||||
return { model, apiKey };
|
||||
}
|
||||
|
||||
private buildContext(messages: Message[], cfg: AgentRunConfig): AgentContext {
|
||||
return {
|
||||
systemPrompt: cfg.systemPrompt,
|
||||
messages,
|
||||
tools: cfg.tools,
|
||||
};
|
||||
}
|
||||
|
||||
const pc: AgentLoopConfig = {
|
||||
private buildLoopConfig(model: typeof cfg.model, apiKey: string, cfg: AgentRunConfig): AgentLoopConfig {
|
||||
return {
|
||||
model,
|
||||
reasoning: cfg.reasoning,
|
||||
apiKey,
|
||||
getQueuedMessages: cfg.getQueuedMessages,
|
||||
};
|
||||
}
|
||||
|
||||
async *run(messages: Message[], userMessage: Message, cfg: AgentRunConfig, signal?: AbortSignal) {
|
||||
const { model, apiKey } = await this.getModelAndKey(cfg);
|
||||
const context = this.buildContext(messages, cfg);
|
||||
const pc = this.buildLoopConfig(model, apiKey, cfg);
|
||||
|
||||
// Yield events from agentLoop
|
||||
for await (const ev of agentLoop(userMessage as unknown as UserMessage, context, pc, signal)) {
|
||||
yield ev;
|
||||
}
|
||||
}
|
||||
|
||||
async *continue(messages: Message[], cfg: AgentRunConfig, signal?: AbortSignal) {
|
||||
const { model, apiKey } = await this.getModelAndKey(cfg);
|
||||
const context = this.buildContext(messages, cfg);
|
||||
const pc = this.buildLoopConfig(model, apiKey, cfg);
|
||||
|
||||
for await (const ev of agentLoopContinue(context, pc, signal)) {
|
||||
yield ev;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -13,10 +13,14 @@ export interface AgentRunConfig {
|
|||
// We re-export the Message type above; consumers should use the upstream AgentEvent type.
|
||||
|
||||
export interface AgentTransport {
|
||||
/** Run with a new user message */
|
||||
run(
|
||||
messages: Message[],
|
||||
userMessage: Message,
|
||||
config: AgentRunConfig,
|
||||
signal?: AbortSignal,
|
||||
): AsyncIterable<AgentEvent>; // passthrough of AgentEvent from upstream
|
||||
): AsyncIterable<AgentEvent>;
|
||||
|
||||
/** Continue from current context (no new user message) */
|
||||
continue(messages: Message[], config: AgentRunConfig, signal?: AbortSignal): AsyncIterable<AgentEvent>;
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue