@autorag/librarian 2.9.0 → 2.10.0

package/README.md

@@ -72,13 +72,13 @@ AutoRAG Agent orchestrates five integrated subsystems:
    - **Datasource Skills:** Federated retrieval across every configured external datasource.
 3. **Result Merger & Scope Narrowing:** Cross-method deduplication, score normalization, and ordinary query-scope narrowing.
 4. **Direct Evidence Reading (`bash`):** The agent directly opens and inspects promising files with `cat`, `grep`, or `find` to verify facts against ground truth.
-5. **Curation:** Structured findings are returned via `emit_autorag_results`; cited evidence that Jev judges to genuinely support the answer is remembered as reference context for later searches ([docs/retrieval-memory.md](docs/retrieval-memory.md)).
+5. **Curation:** The final answer is the model's plain reply; the harness resolves its inline `[eN]` citations into numbered, source-backed results; cited evidence that Jev judges to genuinely support the answer is remembered as reference context for later searches ([docs/retrieval-memory.md](docs/retrieval-memory.md)).
 
 ### Pi host boundary
 
 AutoRAG uses `@earendil-works/pi-coding-agent` as the runtime host for model sessions. Pi owns provider credentials and OAuth storage, model-runtime dispatch, session JSONL persistence/resume, extension loading, lifecycle events, and the built-in `read`/`bash`/`edit`/`write`/`grep`/`find`/`ls` tools. AutoRAG registers only its domain tools and keeps orchestration outside the host: datasource orchestration, MinSync/Jikji preparation, memory hints, fast-to-verification two-phase search, structured result emission, and remote-session filtering.
 
-The CLI TUI (`autorag tui`) is Pi's interactive mode hosted on the AutoRAG librarian: there is no separate AutoRAG renderer. Pi owns the terminal UI and its native in-session commands — `/login`/`/logout`, `/model`, `/resume`, `/new`, `/tree`, `/compact`, and `/settings` — so provider sign-in and model selection work on first launch even when no AutoRAG model is configured. AutoRAG registers its retrieval tools and streams progress, preliminary answers, and final results into the same Pi session, so resume and two-phase search keep working together.
+The CLI TUI (`autorag tui`) is Pi's interactive mode hosted on the AutoRAG librarian: there is no separate AutoRAG renderer. Pi owns the terminal UI and its native in-session commands — `/login`/`/logout`, `/model`, `/resume`, `/new`, `/tree`, `/compact`, and `/settings` — so provider sign-in and model selection work on first launch even when no AutoRAG model is configured. AutoRAG registers its retrieval tools and runs the two-phase search in the same Pi session, where the model's progress notes, first answer, and final answer are ordinary assistant messages, so resume and two-phase search keep working together.
 
 Update notices are AutoRAG's, not Pi's: `autorag tui` suppresses Pi's "run `pi update`" banner (Pi is a bundled host, not a separate install) and instead checks the published `@autorag/librarian` release, showing a one-line notice in the session when a newer version exists. Set `AUTORAG_NO_UPDATE_CHECK=1` to skip the lookup; `AUTORAG_UPDATE_CHECK_URL` overrides the registry endpoint.
 
@@ -359,6 +359,60 @@ for (const result of response.results) {
 
 ---
 
+## Hosted AutoRAG Plan (Dazzi)
+
+AutoRAG can run on the hosted **AutoRAG** plan as a first-class model provider named
+`autorag`. The provider speaks the server's OpenAI Responses API surface (`<root>/v1`),
+so streaming, tool calls, and usage are handled by Pi's built-in `openai-responses`
+implementation, and the model catalog is discovered live from the server — no model id
+is baked into AutoRAG.
+
+Sign in from the interactive TUI:
+
+```text
+autorag tui
+/login          # choose "AutoRAG"; the browser completes a PKCE login
+```
+
+Headless (no browser) with a long-lived API key:
+
+```bash
+export AUTORAG_API_KEY=dz_...
+```
+
+On a fresh agent home, pin the model with an explicit endpoint so it resolves
+offline without a catalog snapshot — this is the headless configuration:
+
+```json
+{
+  "model": {
+    "provider": "autorag",
+    "id": "anthropic/claude-haiku-5.5",
+    "baseUrl": "https://api.dazziapp.com/v1",
+    "api": "openai-responses",
+    "apiKeyEnv": "AUTORAG_API_KEY"
+  }
+}
+```
+
+The `provider`/`id`-only shorthand (`{ "provider": "autorag", "id": "anthropic/claude-haiku-5.5" }`)
+only resolves once a TUI `/login` or a model refresh has persisted the catalog
+snapshot; with no snapshot there is nothing to restore offline, and resolution
+fails with `Unknown configured model: autorag/<id>`.
+
+| Variable | Default | Meaning |
+|---|---|---|
+| `AUTORAG_BASE_URL` | `https://api.dazziapp.com` | Server **root** (not the `/v1` path); the API base is `<root>/v1`. |
+| `AUTORAG_API_KEY` | – | Headless API key. A stored `/login` credential takes precedence. |
+
+The catalog is persisted after a `/login` (or any model refresh), so the
+`provider`/`id` shorthand `model: { provider: "autorag", id }`
+keeps resolving offline. When you are not signed in the provider still registers — it appears under
+`/login`, contributes no models, and makes no network call. See
+[Hosted AutoRAG provider](docs/cloud-provider.md) for the full flow.
+
+---
+
 ## CLI Command Reference
 
 | Command | Description |
@@ -386,6 +440,7 @@ Deep dive into AutoRAG Agent's architecture, security, and integration guides:
 
 - **[MinSync Setup & Embedding QA](docs/minsync-setup.md):** Automatic binary installation, CDC chunking, and EmbeddingGemma verification.
 - **[Local Embedding Runtime & Gateway](docs/embedding-runtime.md):** AutoRAG-owned local gateway, model prefetching, and zero-egress semantic search. Model cards: [`qwen3-embedding-0.6b`](docs/model-cards/qwen3-embedding-0.6b.md), [`embeddinggemma-300m`](docs/model-cards/embeddinggemma-300m.md).
+- **[Hosted AutoRAG Provider](docs/cloud-provider.md):** The `autorag` (Dazzi) model provider — PKCE `/login`, `AUTORAG_API_KEY`, `AUTORAG_BASE_URL`, live model discovery, and offline catalog snapshots.
 - **[Datasource Skills Reference](docs/datasource-skills.md):** Full configuration contracts, connection aliases, and connector options.
 - **[Jev Decisions](docs/jev-decisions.md):** Jev query pipeline, **on by default** via OpenRouter: it routes each question to a direct answer (intrinsic knowledge) or local search (everything needing more information; web search is left to the agent after the fast answer), splits multi-part questions into up to five parallel searches (`openrouter/qwen/qwen3.7-flash`), picks which registered datasources to search before the fast answer, and ends the run after the fast answer when that answer is complete. Also covers the `jev` judgment tool, backends (OpenRouter, TypeSafe, Vercel AI Gateway), and how to opt out.
 - **[Manual QA & Datasource Test Harnesses](docs/manual-qa-datasources.md):** Real-world testing guides for Discord, KakaoTalk, Slack, Notion, and email.

package/dist/agent/agent.d.ts

@@ -17,10 +17,9 @@ import { type DefaultParserRegistryOptions } from "../parser/index.js";
 import { RetrievalEngine } from "../retrieval/engine.js";
 import { RetrievalMethodRegistry } from "../retrieval/registry.js";
 import type { DatasourceCatalogEntry } from "../retrieval/selection.js";
-import type { CuratedResult, RetrievalDiagnostic, RetrievalOptions, RetrievalResult } from "../retrieval/types.js";
+import type { RetrievalDiagnostic, RetrievalOptions, RetrievalResult } from "../retrieval/types.js";
 import { type DatasourceAgentSkill } from "./datasource-skill.js";
 import { type ScanDuplicateDocumentsDetails } from "./dupey-tool.js";
-import { type AutoRAGFastAnswerDetails } from "./fast-answer-tool.js";
 import { type JevToolOptions } from "./jev-extension.js";
 import { type JikjiFindProviderResult } from "./jikji-find-tool.js";
 import { type AutoRAGPiInteractiveRuntime, type AutoRAGPiSession } from "./pi-session.js";
@@ -255,10 +254,10 @@ export interface AutoRAGAgentOptions {
     memoryEmbedder?: Embedder | false;
     /**
      * Agent self-configuration. When set (and Jev is on), Jev gains a `config`
-     * branch for questions about AutoRAG's own settings: the turn skips
-     * `emit_fast_answer`, receives the full `autorag-setup` skill, edits
-     * `configPath` with the pi tools, and reports through
-     * `emit_autorag_results`. Always omitted for remote P2P sessions.
+     * branch for questions about AutoRAG's own settings: the turn skips the
+     * fast answer, receives the full `autorag-setup` skill, edits
+     * `configPath` with the pi tools, and reports what it changed as its final
+     * message. Always omitted for remote P2P sessions.
      */
     selfConfig?: SelfConfigOptions;
     /**
@@ -356,13 +355,10 @@ export declare class AutoRAGAgent {
     private readonly runLogger;
     private lastQuery;
     private lastSessionId;
-    private readonly sessions;
     private activeRun;
-    private resultCapture;
-    private interactiveFastAnswerCallback;
     private modelNativeSearchAuth;
     private retrievalTrace;
-    /** Evidence every retrieval tool returned this run; emit tools resolve model-cited ids against it. */
+    /** Evidence every retrieval tool returned this run; the answer's inline `[eN]` citations resolve against it. */
     private readonly evidenceLedger;
     private preliminaryCallback;
     /** Per-phase thinking levels of the two-phase search. */
@@ -474,6 +470,15 @@ export declare class AutoRAGAgent {
     private currentSystemPromptConfig;
     subscribe(listener: Parameters<Agent["subscribe"]>[0]): () => void;
     createPiInteractiveRuntime(): Promise<AutoRAGPiInteractiveRuntime>;
+    /** Resolve the model's inline `[eN]` citations against this run's evidence and derive the structured results. */
+    private deriveAnswer;
+    /**
+     * Drives one TUI query through the two-phase search. The model's own
+     * assistant messages (progress notes before a tool call, the fast answer,
+     * the final answer) are already recorded and rendered by the bound pi
+     * session, so nothing is re-sent here: a second copy would show every answer
+     * twice and put it in the model's context twice.
+     */
     private runInteractivePiQuery;
     abort(): void;
     /**
@@ -486,13 +491,14 @@ export declare class AutoRAGAgent {
     }): void;
     stopAutoRefresh(): void;
     private runAutoRefreshTick;
-    getResultRegistry(sessionId?: string): ReadonlyMap<number, CuratedResult>;
     searchDocuments(query: string, options?: RetrievalOptions): Promise<SearchDocumentsResponse>;
     /**
      * Stream bounded progress updates while retaining the stable
-     * {@link searchDocuments} promise API. Progress is sourced from model text
-     * deltas, so callers can render it in a TUI, CLI, or parent agent and can
-     * stop by calling {@link abort}.
+     * {@link searchDocuments} promise API. Progress is the text an assistant
+     * message wrote before calling a tool (`stopReason: "toolUse"`); a message
+     * that ends its turn is an answer, delivered as `preliminary` or
+     * `complete`, never as progress. Callers can render it in a TUI, CLI, or
+     * parent agent and can stop by calling {@link abort}.
      */
     searchDocumentsStream(query: string, options?: RetrievalOptions): AsyncGenerator<SearchDocumentsStreamEvent, void, void>;
     /**
@@ -529,7 +535,7 @@ export declare class AutoRAGAgent {
     /** Path-opaque component diagnostics for the search response. */
     private collectComponentDiagnostics;
     /**
-     * Jev check run after emit_fast_answer: does the answer need correction,
+     * Jev check run after the fast answer: does the answer need correction,
      * clarification, or further research? "No" ends the run with the fast
      * answer as the final answer. Without Jev, or when the check fails, the run
      * always continues into verification.
@@ -574,14 +580,14 @@ export declare class AutoRAGAgent {
     private rerankPrefetchPool;
     /**
      * Fast-phase prompt for two-phase searches. The baseline retrieval context
-     * is already gathered, so the model answers immediately with thinking off
-     * via emit_fast_answer, without any further tool calls.
+     * is already gathered, so the model answers immediately with thinking off,
+     * as plain text, without any further tool calls.
      */
     buildFastAnswerPrompt(query: string, options: RetrievalOptions, baseline: string): string;
     /**
      * Prompt for a question Jev routed to a direct answer: general knowledge or
      * small talk. No retrieval ran and no verification phase follows, so the
-     * answer emitted here is final.
+     * reply is the final answer.
      */
     buildDirectAnswerPrompt(query: string): string;
     /** Discovery-tool hint naming only the registered discovery tools; empty when none. */
@@ -595,10 +601,10 @@ export declare class AutoRAGAgent {
     /**
      * Verification-phase prompt for two-phase searches. The fast answer, when a
      * consumer already received it, is embedded verbatim so the model can diff
-     * against it; the model then verifies with thinking on and finalizes with
-     * emit_autorag_results exactly once, returning only the delta.
+     * against it; the model then verifies with thinking on and ends with the
+     * final answer as a plain reply, returning only the delta.
      */
-    buildRefinementPrompt(query: string, options: RetrievalOptions, fastAnswer: AutoRAGFastAnswerDetails | undefined, fastAnswerDelivered: boolean): string;
+    buildRefinementPrompt(query: string, options: RetrievalOptions, fastAnswer: string | undefined, fastAnswerDelivered: boolean): string;
     buildSearchPrompt(query: string, options: RetrievalOptions, initialRetrievalContext?: string): string;
     refresh(force?: boolean, opts?: AutoRAGRefreshOptions): Promise<AutoRAGRefreshResult>;
     /**

package/dist/agent/answer-citations.d.ts

@@ -0,0 +1,22 @@
+import { type EvidenceLedger } from "./evidence-ledger.js";
+import type { AutoRAGResultsDetails } from "./results.js";
+import type { SearchDocumentDiagnostic } from "./search-documents.js";
+export interface DerivedAnswer {
+    readonly details: AutoRAGResultsDetails;
+    readonly diagnostics: readonly SearchDocumentDiagnostic[];
+}
+/**
+ * Turn the model's plain answer into the structured results the rest of the
+ * harness consumes. The model cites with the evidence ids the retrieval tools
+ * printed (`[e3]`, `[e3, e7]`) or `[file:<abs path>]` for a file it opened
+ * itself. Each id is resolved against the run's evidence ledger, so a source,
+ * method, and chunk always come from recorded evidence, never from model text.
+ *
+ * Results are one per cited source, numbered by first appearance, and every
+ * marker in the returned answer is rewritten to its result number (`[1]`).
+ * A marker that matches no evidence, and any numeric marker the model wrote on
+ * its own, is dropped and reported: it can never point at the wrong result.
+ */
+export declare function deriveResultsFromAnswer(text: string, ledger: EvidenceLedger, options: {
+    readonly allowLocalFiles: boolean;
+}): DerivedAnswer;

package/dist/agent/answer-guidelines.d.ts

@@ -1,19 +1,19 @@
 /**
  * The single file-path exception to the answer "no file paths" rule
- * (issue #1790). Every prompt and tool description that shapes `answer` uses
- * this exact wording so a caller that can render images receives a markdown
- * image embed whose path is a real result source, while the no-path rule
- * otherwise still stands.
+ * (issue #1790). Every prompt that shapes the answer uses this exact wording so
+ * a caller that can render images receives a markdown image embed whose path is
+ * a real evidence source, while the no-path rule otherwise still stands.
  */
-export declare const ANSWER_IMAGE_EMBED_RULE = "**Images**: when a retrieved result is itself an image file (png, jpg, jpeg, gif, webp, svg, bmp, avif, heic, tiff) and showing it answers the question, embed it in `answer` as a markdown image on its own line: `![short description](<absolute source path>)`, followed by its citation, e.g. `![Q3 revenue chart](</data/reports/q3 chart.png>) [2]`. Use only the real source path of a result in `results`/`mapping`. Never invent a path, use relative paths, or embed remote URLs. Apart from these image embeds, the no-file-paths rule stands.";
+export declare const ANSWER_IMAGE_EMBED_RULE = "**Images**: when a retrieved result is itself an image file (png, jpg, jpeg, gif, webp, svg, bmp, avif, heic, tiff) and showing it answers the question, embed it in the answer as a markdown image on its own line: `![short description](<absolute source path>)`, followed by its citation, e.g. `![Q3 revenue chart](</data/reports/q3 chart.png>) [e2]`. Use only the real source path of retrieved evidence. Never invent a path, use relative paths, or embed remote URLs. Apart from these image embeds, the no-file-paths rule stands.";
 /**
  * Delta (verification) answers must not re-embed an image the first answer
  * already showed.
  */
 export declare const ANSWER_IMAGE_DELTA_RULE = "An image already embedded in the first answer must not be embedded again.";
 /**
- * One numbering space for citations (issue #1788): every `[n]` in `answer`
- * must be a `results[].number` emitted in the same call. Retrieval candidate
- * numbers and first-answer unit numbers are never citation numbers.
+ * Inline evidence citations. The model writes the id the retrieval tools print
+ * next to each result; the harness resolves it to the recorded source and
+ * renumbers the markers `[1]`, `[2]`, ... in order of first appearance, so the
+ * caller only ever sees numbered citations backed by recorded evidence.
  */
-export declare const ANSWER_CITATION_RULE = "**Citation numbers**: every bracketed citation [n] in `answer` must be the `number` of a result you emit in `results` of the same call. Numbers shown in retrieval candidates or in an earlier first answer are NOT citation numbers \u2014 renumber against your own `results`, and add a result for any evidence you cite.";
+export declare const ANSWER_CITATION_RULE = "**Citations**: cite supporting evidence inline with the evidence id shown next to each retrieved result, in square brackets (e.g. [e3], or [e3][e7] for two). Copy ids exactly; never invent one. For a local file you opened yourself with `bash`, cite `[file:<absolute path>]`. Cite the evidence that backs a sentence at the end of that sentence. Do not cite numbers, paths, or anything else.";

package/dist/agent/citations.d.ts

@@ -1,7 +1,39 @@
-/** Sorted, de-duplicated citation numbers in `answer` with no matching result. */
-export declare function unresolvedCitations(answer: string, results: readonly {
+/**
+ * Bracketed citations in an answer. Two grammars share one scanner:
+ *
+ * - **Final form** `[n]`: a number that must resolve to a `results[].number`
+ *   of the same response (issue #1788).
+ * - **Model form** `[e3]`, `[e3, e7]`, `[file:/abs/path]`: the evidence ids the
+ *   retrieval tools print next to each result, or a local file the model opened
+ *   itself. The harness resolves them against the run's evidence ledger and
+ *   rewrites them into the final form.
+ *
+ * A markdown image target `(<...>)` is skipped verbatim so a bracketed token
+ * inside a real file path is never mistaken for a citation; `[n](...)` is a
+ * markdown link, not a citation.
+ *
+ * `answer` is model- or caller-supplied, so this is a single linear scan rather
+ * than a regex: alternations like `\(<[^>]*>\)` / `[ \t]*\[` backtrack
+ * quadratically on inputs such as repeated `(<` or long whitespace runs.
+ */
+export type AnswerMarker = {
+    readonly kind: "number";
+    readonly start: number;
+    readonly end: number;
     readonly number: number;
-}[]): number[];
+} | {
+    readonly kind: "evidence";
+    readonly start: number;
+    readonly end: number;
+    readonly ids: readonly string[];
+} | {
+    readonly kind: "file";
+    readonly start: number;
+    readonly end: number;
+    readonly path: string;
+};
+/** Every citation marker in `answer`, in order. `isFile` decides which `[file:...]` paths are real. */
+export declare function answerMarkers(answer: string, isFile?: (path: string) => boolean): AnswerMarker[];
 /** Remove citation markers whose number has no matching result. */
 export declare function stripUnresolvedCitations(answer: string, results: readonly {
     readonly number: number;
@@ -10,34 +42,14 @@ export declare function stripUnresolvedCitations(answer: string, results: readon
     readonly unresolved: readonly number[];
 };
 export declare function formatCitationList(numbers: readonly number[]): string;
-/**
- * Throw a corrective error when two results share a number. A repeated number
- * collapses into one registry entry and one feedback id, so the other result's
- * evidence would be lost. Emit tools call this at tool time so the model sees
- * the error and re-emits.
- */
-export declare function assertUniqueResultNumbers(label: string, results: readonly {
-    readonly number: number;
-}[]): void;
 /**
  * Throw a corrective error unless `results` and `mapping` carry the same
  * numbers, exactly one entry each. Duplicates are rejected even when both
  * sides repeat them: a repeated number collapses into a single registry entry,
  * so the other entry's evidence would be lost.
- * Emit tools call this at tool time (issue #1807) so the model sees the error
- * and re-emits; checking only after the run ended failed the whole search and
- * discarded an already-delivered answer.
  */
 export declare function assertResultsMappingOneToOne(label: string, results: readonly {
     readonly number: number;
 }[], mapping: readonly {
     readonly number: number;
 }[]): void;
-/**
- * Throw a corrective error when `answer` cites numbers absent from `results`.
- * Emit tools surface the message to the model as a tool error so it re-emits
- * with one consistent numbering.
- */
-export declare function assertCitationsResolve(label: string, answer: string, results: readonly {
-    readonly number: number;
-}[]): void;

package/dist/agent/evidence-ledger.d.ts

@@ -1,5 +1,5 @@
 import type { RetrievalResult } from "../retrieval/types.js";
-import type { AutoRAGEvidenceRef } from "./emit-results-tool.js";
+import type { AutoRAGEvidenceRef } from "./results.js";
 export interface LedgerEvidenceInput {
     readonly method: string;
     readonly source: string;
@@ -13,16 +13,11 @@ export interface LedgerEvidenceInput {
     readonly evidenceType?: string;
     readonly evidenceLocation?: string;
 }
-export interface ResolveEvidenceOptions {
-    /** Tool name used to prefix corrective errors. */
-    readonly label: string;
-    /** Result number the refs belong to; named in corrective errors. */
-    readonly number: number;
-    /** Excerpt recorded for a local file the model opened itself (no search returned it). */
-    readonly fallbackContent: string;
+export interface LookupEvidenceOptions {
     /** False in remote sessions: only evidence a tool returned this run may be cited. */
     readonly allowLocalFiles: boolean;
 }
+export declare function isRegularFile(path: string): boolean;
 /**
  * Per-run record of every piece of evidence a retrieval tool returned.
  *
@@ -47,13 +42,13 @@ export declare class EvidenceLedger {
     /** Record one retrieval result as returned by `tool`; backend metadata is carried over. */
     registerResult(tool: string, result: RetrievalResult): string;
     /**
-     * Resolve model-supplied refs to recorded evidence. A ref is an evidence id
+     * Resolve one cited reference to recorded evidence. A ref is an evidence id
      * (`e3` or `[e3]`), the exact source a tool returned (a URL, a datasource
      * id, a path), or the absolute path of a real local file the model opened
-     * itself. Anything else throws a corrective error naming the result.
+     * itself. Returns an empty list when nothing matches: the caller drops the
+     * citation instead of failing the run.
      */
-    resolve(refs: readonly string[], options: ResolveEvidenceOptions): AutoRAGEvidenceRef[];
+    lookup(raw: string, options: LookupEvidenceOptions): AutoRAGEvidenceRef[];
     /** Forget every recorded entry. Ids are never reissued, so a previous run's id stays unresolvable. */
     clear(): void;
-    private unresolvedMessage;
 }

package/dist/agent/index.d.ts

@@ -2,11 +2,11 @@ export type { AutoRAGAgentOptions, AutoRefreshOptions, PeerQueryOptions } from "
 export { AutoRAGAgent } from "./agent.js";
 export { buildDatasourceSkillsPrompt, createLoadDatasourceSkillTool, type DatasourceSkillProvider, datasourceSkillLocation, formatDatasourceSkillInvocation, LOAD_DATASOURCE_SKILL_TOOL_NAME, type LoadDatasourceSkillDetails, toDatasourceAgentSkill, } from "./datasource-skill.js";
 export { createScanDuplicateDocumentsTool, SCAN_DUPLICATE_DOCUMENTS_TOOL_NAME, type ScanDuplicateDocumentsDetails, type ScanDuplicateDocumentsProvider, } from "./dupey-tool.js";
-export type { AutoRAGEmittedResult, AutoRAGMappingEntry, AutoRAGResultsDetails, } from "./emit-results-tool.js";
-export { createEmitResultsTool, EMIT_AUTORAG_RESULTS_TOOL_NAME, emitResultsSchema, reportSchema, } from "./emit-results-tool.js";
 export { createJevExtension, JEV_TOOL_NAME, type JevBackendName, type JevToolOptions, } from "./jev-extension.js";
 export { createJikjiFindTool, JIKJI_FIND_TOOL_NAME, type JikjiFindDetails, type JikjiFindPerRootPolicy, type JikjiFindProvider, type JikjiFindProviderResult, type MergedJikjiPolicy, } from "./jikji-find-tool.js";
 export { createQueryPeerAgentTool, QUERY_PEER_AGENT_TOOL_NAME, type QueryPeerAgentDetails, type QueryPeerAgentResult, type QueryPeerAgentToolOptions, } from "./query-peer-tool.js";
+export type { AutoRAGEmittedResult, AutoRAGMappingEntry, AutoRAGResultsDetails, } from "./results.js";
+export { reportSchema } from "./results.js";
 export { createSearchAllDocumentsTool, SEARCH_ALL_DOCUMENTS_TOOL_NAME, type SearchAllDocumentsDetails, type SearchAllDocumentsProvider, type SearchAllDocumentsResult, } from "./search-all-tool.js";
 export type { SearchDocumentDiagnostic, SearchDocumentDiagnosticCode, SearchDocumentDiagnosticSeverity, SearchDocumentEvidence, SearchDocumentResult, SearchDocumentsResponse, SearchDocumentsStreamEvent, SearchDocumentWarning, } from "./search-documents.js";
 export { createSearchMinSyncDocumentsTool, SEARCH_MINSYNC_DOCUMENTS_TOOL_NAME, type SearchMinSyncDocumentsDetails, } from "./search-minsync-tool.js";

package/dist/agent/{emit-results-tool.d.ts → results.d.ts}

@@ -1,33 +1,11 @@
-import type { AgentTool } from "@earendil-works/pi-agent-core";
 import { Type } from "typebox";
-import type { EvidenceLedger } from "./evidence-ledger.js";
-export declare const EMIT_AUTORAG_RESULTS_TOOL_NAME = "emit_autorag_results";
-/**
- * Model-facing schema of `emit_autorag_results`. The model cites evidence by
- * the short ids retrieval tools print next to each result (`refs`); the
- * harness resolves them to the recorded source, method, and chunk, so the
- * model never retypes a path or a chunk.
- */
-export declare const emitResultsSchema: Type.TObject<{
-    answer: Type.TString;
-    results: Type.TArray<Type.TObject<{
-        number: Type.TInteger;
-        title: Type.TString;
-        summary: Type.TString;
-        evidence: Type.TArray<Type.TObject<{
-            excerpt: Type.TString;
-            lineNumber: Type.TOptional<Type.TInteger>;
-        }>>;
-        confidence: Type.TNumber;
-        refs: Type.TArray<Type.TString>;
-    }>>;
-    warnings: Type.TOptional<Type.TArray<Type.TString>>;
-}>;
 /**
  * Report schema with the explicit number -> source mapping. This is the
  * persisted/typed shape and the input contract of `autorag report` and the MCP
  * report tool, where an external agent curated the evidence and no harness
- * ledger exists.
+ * ledger exists. The librarian itself never produces this shape from the
+ * model: it writes a plain answer and the harness derives the results from the
+ * evidence ids the answer cites.
  */
 export declare const reportSchema: Type.TObject<{
     answer: Type.TString;
@@ -75,7 +53,8 @@ export interface AutoRAGEmittedResult {
     readonly title: string;
     readonly summary: string;
     readonly evidence: readonly AutoRAGEmittedEvidence[];
-    readonly confidence: number;
+    /** Absent for results the harness derived from cited evidence: nothing measured a confidence. */
+    readonly confidence?: number;
 }
 export interface AutoRAGEvidenceRef {
     readonly method: string;
@@ -107,17 +86,3 @@ export interface AutoRAGResultsDetails {
     readonly mapping: readonly AutoRAGMappingEntry[];
     readonly warnings: readonly string[];
 }
-export interface EmitResultsToolOptions {
-    /** Evidence the run's retrieval tools returned; `refs` resolve against it. */
-    readonly ledger: EvidenceLedger;
-    /** False in remote sessions: only evidence a tool returned this run may be cited. */
-    readonly allowLocalFiles: boolean;
-}
-/**
- * Builds the terminating structured-result tool. The model calls this exactly
- * once as its final action; the typed `details` plus `terminate: true` end the
- * Pi Agent run and hand the curated results back through `capture` — no
- * assistant-text parsing involved. The number -> source mapping is built here
- * from the harness-held evidence the model cited by id, never from model text.
- */
-export declare function createEmitResultsTool(capture: (details: AutoRAGResultsDetails) => void, options: EmitResultsToolOptions): AgentTool<typeof emitResultsSchema, AutoRAGResultsDetails>;

package/dist/agent/search-documents.d.ts

@@ -1,7 +1,6 @@
-import { type RetrievalMemory } from "../memory/memory.js";
-import type { CuratedResult, RetrievalResult } from "../retrieval/types.js";
-import type { AutoRAGResultsDetails } from "./emit-results-tool.js";
-import type { AutoRAGFastAnswerDetails } from "./fast-answer-tool.js";
+import { type RetrievalMemory, type SessionEvidenceRef } from "../memory/memory.js";
+import type { RetrievalResult } from "../retrieval/types.js";
+import type { AutoRAGMappingEntry, AutoRAGResultsDetails } from "./results.js";
 export type SearchDocumentWarning = "empty-query";
 export type SearchDocumentDiagnosticSeverity = "info" | "warning" | "error";
 /**
@@ -10,7 +9,7 @@ export type SearchDocumentDiagnosticSeverity = "info" | "warning" | "error";
  * (#21/#22) does not require another source-breaking type change; codes not yet
  * emitted are still valid members.
  */
-export type SearchDocumentDiagnosticCode = "empty-query" | "no-verified-results" | "unknown-warning" | "caller-tool-dropped" | "minsync-staging-excluded" | "minsync-sync-failed" | "embedder-unavailable" | "embedding-identity-mismatch" | "parser-skipped" | "parser-failed" | "source-vanished" | "source-unreadable" | "parser-warning" | "duplicate-excluded" | "unsupported-file" | "stale-index" | "deleted-mirror" | "pdf-extract-thin" | "retrieval-method-failed" | "rerank-failed" | "jikji-unavailable" | "jikji-prepare-failed" | "jikji-find-failed" | "everything-index-failed" | "fsearch-index-failed" | "fsearch-binary-missing" | "refresh-failed" | "refresh-interrupted" | "watch-failed" | "watch-limited" | "unknown-datasource-skill" | "missing-final-emit" | "model-request-failed" | "search-timeout" | "query-routed" | "query-route-fallback" | "query-decomposition-failed" | "follow-up-skipped" | "follow-up-check-fallback" | "datasources-selected" | "datasource-selection-fallback" | "citation-without-result" | "evidence-judged" | "evidence-judgment-fallback" | "memory-search-fallback" | "self-config-unavailable" | "self-config-rolled-back";
+export type SearchDocumentDiagnosticCode = "empty-query" | "no-verified-results" | "unknown-warning" | "caller-tool-dropped" | "minsync-staging-excluded" | "minsync-sync-failed" | "embedder-unavailable" | "embedding-identity-mismatch" | "parser-skipped" | "parser-failed" | "source-vanished" | "source-unreadable" | "parser-warning" | "duplicate-excluded" | "unsupported-file" | "stale-index" | "deleted-mirror" | "pdf-extract-thin" | "retrieval-method-failed" | "rerank-failed" | "jikji-unavailable" | "jikji-prepare-failed" | "jikji-find-failed" | "everything-index-failed" | "fsearch-index-failed" | "fsearch-binary-missing" | "refresh-failed" | "refresh-interrupted" | "watch-failed" | "watch-limited" | "unknown-datasource-skill" | "no-final-answer" | "model-request-failed" | "search-timeout" | "query-routed" | "query-route-fallback" | "query-decomposition-failed" | "follow-up-skipped" | "follow-up-check-fallback" | "datasources-selected" | "datasource-selection-fallback" | "citation-without-result" | "evidence-judged" | "evidence-judgment-fallback" | "memory-search-fallback" | "self-config-unavailable" | "self-config-rolled-back";
 export interface SearchDocumentDiagnostic {
     readonly code: SearchDocumentDiagnosticCode;
     readonly severity: SearchDocumentDiagnosticSeverity;
@@ -27,7 +26,8 @@ export interface SearchDocumentResult {
     readonly title: string;
     readonly summary: string;
     readonly evidence: readonly SearchDocumentEvidence[];
-    readonly confidence: number;
+    /** Absent when nothing measured one: derived results carry no model-reported confidence. */
+    readonly confidence?: number;
     readonly source?: string;
 }
 /** One retrieval candidate captured in a run's retrieval trace. */
@@ -38,7 +38,7 @@ export interface SearchDocumentRetrievalTraceResult {
 }
 /**
  * What one retrieval tool execution found during a search run. Attached to the
- * degraded fallback response when the agent never called emit_autorag_results,
+ * degraded fallback response when the agent never wrote a final answer,
  * so an upstream agent can still inspect the raw candidates.
  */
 export interface SearchDocumentRetrievalTraceEntry {
@@ -84,20 +84,15 @@ export type SearchDocumentsStreamEvent = {
     readonly type: "complete";
     readonly response: SearchDocumentsResponse;
 };
-type SearchSession = {
-    query: string;
-    registry: Map<number, CuratedResult>;
-};
-type SearchSessions = Map<string, SearchSession>;
 /**
  * Build the preliminary (fast-phase) search response. Unlike
- * {@link recordStructuredResultsSession} this NEVER touches memory or the
- * session registry — the final response owns those.
+ * {@link recordStructuredResultsSession} this NEVER touches memory — the final
+ * response owns that.
  */
-export declare function createPreliminarySearchDocumentsResponse(sessionId: string, query: string, details: AutoRAGFastAnswerDetails, diagnostics?: readonly SearchDocumentDiagnostic[]): SearchDocumentsResponse;
-export declare function createEmptySearchDocumentsResponse(sessionId: string, query: string, sessions: SearchSessions, diagnostics?: readonly SearchDocumentDiagnostic[]): SearchDocumentsResponse;
-export declare function recordStructuredResultsSession(sessionId: string, query: string, details: AutoRAGResultsDetails, sessions: SearchSessions, memory: RetrievalMemory, componentDiagnostics?: readonly SearchDocumentDiagnostic[], options?: {
-    /** Record the session registry only, skipping the retrieval-memory write. */
+export declare function createPreliminarySearchDocumentsResponse(sessionId: string, query: string, details: AutoRAGResultsDetails, diagnostics?: readonly SearchDocumentDiagnostic[]): SearchDocumentsResponse;
+export declare function createEmptySearchDocumentsResponse(sessionId: string, query: string, diagnostics?: readonly SearchDocumentDiagnostic[]): SearchDocumentsResponse;
+export declare function normalizeEntryEvidenceRefs(entry: AutoRAGMappingEntry): SessionEvidenceRef[];
+export declare function recordStructuredResultsSession(sessionId: string, query: string, details: AutoRAGResultsDetails, memory: RetrievalMemory, componentDiagnostics?: readonly SearchDocumentDiagnostic[], options?: {
+    /** Skip the retrieval-memory write. */
     readonly isolateMemory?: boolean;
 }): SearchDocumentsResponse;
-export {};

package/dist/agent/self-config.d.ts

@@ -35,6 +35,6 @@ export declare function rollbackIfBroken(options: SelfConfigOptions, before: str
 /**
  * The turn prompt for the `config` branch: the whole setup skill, where this
  * run's config lives, the safety rules for editing it, and the user's request.
- * The turn ends with `emit_autorag_results`; there is no fast answer.
+ * The turn ends with the model's plain final message; there is no fast answer.
  */
 export declare function buildSelfConfigPrompt(input: SelfConfigPromptInput): string;

package/dist/cli/commands/report.d.ts

@@ -1,4 +1,4 @@
-import { type AutoRAGResultsDetails } from "../../agent/emit-results-tool.js";
+import { type AutoRAGResultsDetails } from "../../agent/results.js";
 import type { CommandContext } from "./types.js";
 /** Validate a curated report payload against the `autorag report` contract (explicit number -> source mapping). */
 export declare function validateReport(value: unknown): AutoRAGResultsDetails;

package/dist/cli/commands/tui.d.ts

@@ -5,8 +5,9 @@ import type { CommandContext } from "./types.js";
  * Pi owns the terminal UI, `read`/`bash`/`edit`/`write`/`grep`/`find`/`ls`
  * tools, provider credentials/OAuth, model selection, session persistence and
  * resume/fork/compaction, settings, and extension loading. AutoRAG registers
- * its domain tools and streams progress, preliminary answers, and final results
- * into the same session through the agent's interactive runtime. The native
+ * its domain tools and runs the two-phase search in the same session through
+ * the agent's interactive runtime; the model's progress notes, first answer, and
+ * final answer are ordinary assistant messages there. The native
  * `/login`, `/model`, `/resume`, `/new`, `/tree`, `/compact`, and `/settings`
  * commands are therefore always available.
  */