mirror of
https://github.com/getcompanion-ai/co-mono.git
synced 2026-04-15 11:02:17 +00:00
Add full-compaction hook example
This commit is contained in:
Mario Zechner
parent
bf313d2713
commit
5aa53ded11
3 changed files with 143 additions and 51 deletions
|
|
@ -6359,6 +6359,23 @@ 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",
|
||||
|
|
@ -6393,23 +6410,6 @@ 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",
|
||||
|
|
@ -6546,23 +6546,6 @@ 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",
|
||||
|
|
@ -6597,6 +6580,23 @@ 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-3.5-turbo-0613": {
|
||||
id: "openai/gpt-3.5-turbo-0613",
|
||||
name: "OpenAI: GPT-3.5 Turbo (older v0613)",
|
||||
api: "openai-completions",
|
||||
provider: "openrouter",
|
||||
baseUrl: "https://openrouter.ai/api/v1",
|
||||
reasoning: false,
|
||||
input: ["text"],
|
||||
cost: {
|
||||
input: 1,
|
||||
output: 2,
|
||||
cacheRead: 0,
|
||||
cacheWrite: 0,
|
||||
},
|
||||
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",
|
||||
|
|
@ -6750,6 +6733,23 @@ export const MODELS = {
|
|||
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)",
|
||||
api: "openai-completions",
|
||||
provider: "openrouter",
|
||||
baseUrl: "https://openrouter.ai/api/v1",
|
||||
reasoning: false,
|
||||
input: ["text"],
|
||||
cost: {
|
||||
input: 1,
|
||||
output: 2,
|
||||
cacheRead: 0,
|
||||
cacheWrite: 0,
|
||||
},
|
||||
contextWindow: 4095,
|
||||
maxTokens: 4096,
|
||||
} satisfies Model<"openai-completions">,
|
||||
"mistralai/mistral-tiny": {
|
||||
id: "mistralai/mistral-tiny",
|
||||
name: "Mistral Tiny",
|
||||
|
|
|
|||
|
|
@ -25,6 +25,9 @@ Prevents session changes when there are uncommitted git changes. Blocks clear/sw
|
|||
### auto-commit-on-exit.ts
|
||||
Automatically commits changes when the agent exits (shutdown event). Uses the last assistant message to generate a commit message.
|
||||
|
||||
### full-compaction.ts
|
||||
Custom context compaction that summarizes the entire conversation instead of keeping recent turns. Uses the `before_compact` hook event to intercept compaction and generate a comprehensive summary using `complete()` from the AI package. Useful when you want maximum context window space at the cost of losing exact conversation history.
|
||||
|
||||
## Usage
|
||||
|
||||
```bash
|
||||
|
|
|
|||
89
packages/coding-agent/examples/hooks/full-compaction.ts
Normal file
89
packages/coding-agent/examples/hooks/full-compaction.ts
Normal file
|
|
@ -0,0 +1,89 @@
|
|||
/**
|
||||
* Full Context Compaction Hook
|
||||
*
|
||||
* Replaces the default compaction behavior with a full summary of the entire context.
|
||||
* Instead of keeping the last 20k tokens of conversation turns, this hook:
|
||||
* 1. Summarizes ALL messages being compacted into a single comprehensive summary
|
||||
* 2. Discards all old turns completely
|
||||
*
|
||||
* This is useful when you want maximum context window space for new work
|
||||
* at the cost of losing exact conversation history.
|
||||
*
|
||||
* Usage:
|
||||
* pi --hook examples/hooks/full-compaction.ts
|
||||
*/
|
||||
|
||||
import { complete } from "@mariozechner/pi-ai";
|
||||
import { messageTransformer } from "@mariozechner/pi-coding-agent";
|
||||
import type { HookAPI } from "@mariozechner/pi-coding-agent/hooks";
|
||||
|
||||
export default function (pi: HookAPI) {
|
||||
pi.on("session", async (event, ctx) => {
|
||||
if (event.reason !== "before_compact") return;
|
||||
|
||||
const { messagesToSummarize, tokensBefore, model, apiKey, cutPoint } = event;
|
||||
|
||||
ctx.ui.notify(`Compacting ${tokensBefore.toLocaleString()} tokens with full summary...`, "info");
|
||||
|
||||
// Transform app messages to LLM-compatible format
|
||||
const transformedMessages = messageTransformer(messagesToSummarize);
|
||||
|
||||
// Build messages that ask for a comprehensive summary
|
||||
const summaryMessages = [
|
||||
...transformedMessages,
|
||||
{
|
||||
role: "user" as const,
|
||||
content: [
|
||||
{
|
||||
type: "text" as const,
|
||||
text: `You are a conversation summarizer. Create a comprehensive summary of this conversation that captures:
|
||||
|
||||
1. The main goals and objectives discussed
|
||||
2. Key decisions made and their rationale
|
||||
3. Important code changes, file modifications, or technical details
|
||||
4. Current state of any ongoing work
|
||||
5. Any blockers, issues, or open questions
|
||||
6. Next steps that were planned or suggested
|
||||
|
||||
Be thorough but concise. The summary will replace the entire conversation history, so include all information needed to continue the work effectively.
|
||||
|
||||
Format the summary as structured markdown with clear sections.`,
|
||||
},
|
||||
],
|
||||
timestamp: Date.now(),
|
||||
},
|
||||
];
|
||||
|
||||
try {
|
||||
// Use the same model and API key that would be used for compaction
|
||||
const response = await complete(model, { messages: summaryMessages }, { apiKey, maxTokens: 8192 });
|
||||
|
||||
const summary = response.content
|
||||
.filter((c): c is { type: "text"; text: string } => c.type === "text")
|
||||
.map((c) => c.text)
|
||||
.join("\n");
|
||||
|
||||
if (!summary.trim()) {
|
||||
ctx.ui.notify("Compaction summary was empty, using default compaction", "warning");
|
||||
return; // Fall back to default compaction
|
||||
}
|
||||
|
||||
// Return a compaction entry that discards ALL old messages
|
||||
// firstKeptEntryIndex points to after all summarized content
|
||||
return {
|
||||
compactionEntry: {
|
||||
type: "compaction" as const,
|
||||
timestamp: new Date().toISOString(),
|
||||
summary,
|
||||
firstKeptEntryIndex: cutPoint.firstKeptEntryIndex,
|
||||
tokensBefore,
|
||||
},
|
||||
};
|
||||
} catch (error) {
|
||||
const message = error instanceof Error ? error.message : String(error);
|
||||
ctx.ui.notify(`Compaction failed: ${message}`, "error");
|
||||
// Fall back to default compaction on error
|
||||
return;
|
||||
}
|
||||
});
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue