@caupulican/pi-adaptative 0.80.73 → 0.80.75

package/dist/core/agent-session.js

@@ -39,7 +39,7 @@ import { FileStoreProvider } from "./memory/providers/file-store.js";
 import { TranscriptRecallProvider } from "./memory/providers/transcript-recall.js";
 import { compactToolResultDetailsForRetention } from "./message-retention.js";
 import { createCustomMessage } from "./messages.js";
-import { resolveProfileModelSettings } from "./model-resolver.js";
+import { resolveCliModel, resolveProfileModelSettings } from "./model-resolver.js";
 import { expandPromptTemplate } from "./prompt-templates.js";
 import { stripResourceProfileBlocks } from "./resource-profile-blocks.js";
 import { classifyToolTrust, UNTRUSTED_BOUNDARY_SYSTEM_RULE, wrapUntrustedText } from "./security/untrusted-boundary.js";
@@ -219,6 +219,39 @@ export class AgentSession {
         const result = await this._modelRegistry.getApiKeyAndHeaders(model);
         return result.ok ? { apiKey: result.apiKey, headers: result.headers } : {};
     }
+    /**
+     * Resolve the model used to SUMMARIZE during compaction (cost guard, #30). A compaction summary is an
+     * extraction task — it does not need the main (expensive) model. Selection:
+     *   - an explicit `compaction.model` setting wins, but only if its provider is authed (else fall back);
+     *   - `"auto"` (default) picks the CHEAPEST authed model whose context window can hold a compaction
+     *     (capability floor), and ONLY if it is strictly cheaper than the session model — so we never
+     *     downgrade to an equally-priced but weaker summarizer (agy's floor: don't degrade the checkpoint);
+     *   - otherwise the session model is used (safe default).
+     */
+    _resolveCompactionModel(sessionModel) {
+        const setting = this.settingsManager.getCompactionModel();
+        if (setting && setting !== "auto") {
+            const resolved = resolveCliModel({ cliModel: setting, modelRegistry: this._modelRegistry });
+            if (resolved.model && this._modelRegistry.hasConfiguredAuth(resolved.model))
+                return resolved.model;
+            return sessionModel; // configured but unusable → don't break compaction
+        }
+        // "auto": cheapest authed model that can summarize a large context AND is cheaper than the session
+        // model. The context-window floor keeps a tiny local model from being picked for a big summary.
+        const FLOOR_CONTEXT = 64_000;
+        const sessionInputCost = sessionModel.cost?.input ?? Number.POSITIVE_INFINITY;
+        let best;
+        for (const m of this._modelRegistry.getAvailable()) {
+            if ((m.contextWindow ?? 0) < FLOOR_CONTEXT)
+                continue;
+            const cost = m.cost?.input ?? Number.POSITIVE_INFINITY;
+            if (cost >= sessionInputCost)
+                continue; // only ever pick something cheaper than the session model
+            if (!best || cost < (best.cost?.input ?? Number.POSITIVE_INFINITY))
+                best = m;
+        }
+        return best ?? sessionModel;
+    }
     /**
      * Install tool hooks once on the Agent instance.
      *
@@ -1676,7 +1709,8 @@ export class AgentSession {
             if (!this.model) {
                 throw new Error(formatNoModelSelectedMessage());
             }
-            const { apiKey, headers } = await this._getCompactionRequestAuth(this.model);
+            const compactionModel = this._resolveCompactionModel(this.model);
+            const { apiKey, headers } = await this._getCompactionRequestAuth(compactionModel);
             const pathEntries = this.sessionManager.getBranch();
             const settings = this.settingsManager.getCompactionSettings();
             const preparation = prepareCompaction(pathEntries, settings);
@@ -1719,7 +1753,7 @@ export class AgentSession {
             }
             else {
                 // Generate compaction result
-                const result = await compact(preparation, this.model, apiKey, headers, customInstructions, this._compactionAbortController.signal, this.thinkingLevel, this.agent.streamFn);
+                const result = await compact(preparation, compactionModel, apiKey, headers, customInstructions, this._compactionAbortController.signal, this.thinkingLevel, this.agent.streamFn);
                 summary = result.summary;
                 firstKeptEntryId = result.firstKeptEntryId;
                 tokensBefore = result.tokensBefore;
@@ -1897,10 +1931,12 @@ export class AgentSession {
                 });
                 return false;
             }
+            // Summarize with the cheap auxiliary model when available (cost guard, #30).
+            const compactionModel = this._resolveCompactionModel(this.model);
             let apiKey;
             let headers;
             if (this.agent.streamFn === streamSimple) {
-                const authResult = await this._modelRegistry.getApiKeyAndHeaders(this.model);
+                const authResult = await this._modelRegistry.getApiKeyAndHeaders(compactionModel);
                 if (!authResult.ok || !authResult.apiKey) {
                     this._emit({
                         type: "compaction_end",
@@ -1915,7 +1951,7 @@ export class AgentSession {
                 headers = authResult.headers;
             }
             else {
-                ({ apiKey, headers } = await this._getCompactionRequestAuth(this.model));
+                ({ apiKey, headers } = await this._getCompactionRequestAuth(compactionModel));
             }
             const pathEntries = this.sessionManager.getBranch();
             const preparation = prepareCompaction(pathEntries, settings);
@@ -1967,7 +2003,7 @@ export class AgentSession {
             }
             else {
                 // Generate compaction result
-                const compactResult = await compact(preparation, this.model, apiKey, headers, undefined, this._autoCompactionAbortController.signal, this.thinkingLevel, this.agent.streamFn);
+                const compactResult = await compact(preparation, compactionModel, apiKey, headers, undefined, this._autoCompactionAbortController.signal, this.thinkingLevel, this.agent.streamFn);
                 summary = compactResult.summary;
                 firstKeptEntryId = compactResult.firstKeptEntryId;
                 tokensBefore = compactResult.tokensBefore;