ollama-intern-mcp 1.0.0

package/dist/index.js

@@ -0,0 +1,153 @@
+#!/usr/bin/env node
+/**
+ * Ollama Intern MCP — entrypoint.
+ *
+ * Registers the 8-tool labor surface. Tool descriptions encode *when* to reach
+ * for them — Claude picks the tier by picking the tool, and the tool implies
+ * the model. ollama_chat is visibly last-resort.
+ */
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
+import { fileURLToPath } from "node:url";
+import { realpathSync } from "node:fs";
+import { VERSION } from "./version.js";
+import { loadProfile } from "./profiles.js";
+import { HttpOllamaClient } from "./ollama.js";
+import { NdjsonLogger } from "./observability.js";
+import { toErrorShape } from "./errors.js";
+import { runPrewarm } from "./prewarm.js";
+import { classifySchema, handleClassify } from "./tools/classify.js";
+import { triageLogsSchema, handleTriageLogs } from "./tools/triageLogs.js";
+import { summarizeFastSchema, handleSummarizeFast } from "./tools/summarizeFast.js";
+import { summarizeDeepSchema, handleSummarizeDeep } from "./tools/summarizeDeep.js";
+import { draftSchema, handleDraft } from "./tools/draft.js";
+import { extractSchema, handleExtract } from "./tools/extract.js";
+import { researchSchema, handleResearch } from "./tools/research.js";
+import { embedSchema, handleEmbed } from "./tools/embed.js";
+import { embedSearchSchema, handleEmbedSearch } from "./tools/embedSearch.js";
+import { corpusIndexSchema, handleCorpusIndex } from "./tools/corpusIndex.js";
+import { corpusSearchSchema, handleCorpusSearch } from "./tools/corpusSearch.js";
+import { corpusAnswerSchema, handleCorpusAnswer } from "./tools/corpusAnswer.js";
+import { corpusRefreshSchema, handleCorpusRefresh } from "./tools/corpusRefresh.js";
+import { corpusListSchema, handleCorpusList } from "./tools/corpusList.js";
+import { incidentBriefSchema, handleIncidentBrief } from "./tools/incidentBrief.js";
+import { repoBriefSchema, handleRepoBrief } from "./tools/repoBrief.js";
+import { changeBriefSchema, handleChangeBrief } from "./tools/changeBrief.js";
+import { incidentPackSchema, handleIncidentPack } from "./tools/packs/incidentPack.js";
+import { repoPackSchema, handleRepoPack } from "./tools/packs/repoPack.js";
+import { changePackSchema, handleChangePack } from "./tools/packs/changePack.js";
+import { artifactListSchema, handleArtifactList } from "./tools/artifactList.js";
+import { artifactReadSchema, handleArtifactRead } from "./tools/artifactRead.js";
+import { artifactDiffSchema, handleArtifactDiff } from "./tools/artifactDiff.js";
+import { artifactExportToPathSchema, handleArtifactExportToPath } from "./tools/artifactExportToPath.js";
+import { artifactIncidentNoteSnippetSchema, handleArtifactIncidentNoteSnippet, artifactOnboardingSectionSnippetSchema, handleArtifactOnboardingSectionSnippet, artifactReleaseNoteSnippetSchema, handleArtifactReleaseNoteSnippet, } from "./tools/artifactSnippets.js";
+import { chatSchema, handleChat } from "./tools/chat.js";
+export function createServer(ctx) {
+    const server = new McpServer({ name: "ollama-intern-mcp", version: VERSION });
+    const wrap = (p) => p.then((value) => ({ content: [{ type: "text", text: JSON.stringify(value, null, 2) }] }), (err) => ({
+        content: [{ type: "text", text: JSON.stringify(toErrorShape(err), null, 2) }],
+        isError: true,
+    }));
+    // FLAGSHIP — ollama_research
+    server.tool("ollama_research", "FLAGSHIP. Answer a question grounded in specific files. Takes FILE PATHS (not raw text) — reads and chunks locally, returns a digest with validated citations. Use this to understand a repo/doc without burning Claude context on the full content. Citations outside source_paths are stripped server-side.", researchSchema.shape, (args) => wrap(handleResearch(args, ctx)));
+    // FLAGSHIP — ollama_corpus_search (persistent concept search over named corpora)
+    server.tool("ollama_corpus_search", "FLAGSHIP. Concept search over a persistent named corpus (e.g. 'memory', 'canon', 'handbook'). Pass `corpus` + `query`; returns ranked `[{id, path, score, chunk_index, preview?}]` drawn from the indexed corpus. Use this as your default for semantic recall — the corpus is persistent across sessions so you don't re-embed every call. Build or refresh a corpus with ollama_corpus_index first; see what's available with ollama_corpus_list.", corpusSearchSchema.shape, (args) => wrap(handleCorpusSearch(args, ctx)));
+    // FLAGSHIP — ollama_corpus_answer (grounded synthesis over a named corpus)
+    server.tool("ollama_corpus_answer", "FLAGSHIP. Answer a question from a NAMED CORPUS with chunk-grounded citations. Retrieves via corpus_search, synthesizes with the Deep tier from the retrieved chunks ONLY, and returns `{answer, citations:[{path, chunk_index, heading_path, title}], covered_sources, omitted_sources, coverage_notes, retrieval:{retrieved, top_score, weak}}`. Distinct from ollama_research: research takes source paths you explicitly hand in; corpus_answer pulls from an already-indexed corpus. Weak retrieval degrades honestly — 0 hits short-circuits without invoking the model; thin retrieval flags `weak: true`.", corpusAnswerSchema.shape, (args) => wrap(handleCorpusAnswer(args, ctx)));
+    // FLAGSHIP — ollama_incident_brief (structured operator brief)
+    server.tool("ollama_incident_brief", "FLAGSHIP compound job. Produces a STRUCTURED OPERATOR BRIEF from log_text and/or source_paths, optionally blended with a named corpus for background context. Returns `{root_cause_hypotheses, affected_surfaces, timeline_clues, next_checks, evidence, weak, coverage_notes, corpus_used}`. Every hypothesis/surface/clue carries evidence_refs into the evidence array — refs to unknown ids are stripped server-side. Distinct from ollama_triage_logs (symptoms in one blob) and ollama_research (answer a specific question). Thin evidence degrades to weak=true with coverage_notes — never a smooth fake narrative. next_checks are INVESTIGATIVE, not remediations.", incidentBriefSchema.shape, (args) => wrap(handleIncidentBrief(args, ctx)));
+    // FLAGSHIP — ollama_repo_brief (operator map of a repo)
+    server.tool("ollama_repo_brief", "FLAGSHIP compound job. Produces an OPERATOR MAP of a repo: `{repo_thesis, key_surfaces, architecture_shape, risk_areas, read_next, evidence, weak, coverage_notes, corpus_used}`. Takes source_paths (typically README + key src entries + manifests + docs) and optionally a corpus for cross-cutting context. Not a research clone — research answers a specific question; repo_brief synthesizes orientation. Every key_surface and risk_area cites evidence. read_next is INVESTIGATIVE (files or sections to look at), never prescriptive fixes or refactors. Thin evidence → weak=true with coverage notes.", repoBriefSchema.shape, (args) => wrap(handleRepoBrief(args, ctx)));
+    // FLAGSHIP — ollama_change_brief (structured impact brief for a change)
+    server.tool("ollama_change_brief", "FLAGSHIP compound job. Produces a CHANGE IMPACT BRIEF: `{change_summary, affected_surfaces, why_it_matters, likely_breakpoints, validation_checks, release_note_draft, evidence, weak, coverage_notes, corpus_used}`. Accepts diff_text (split per file on `diff --git` markers) and/or source_paths (changed files), with an optional corpus for architecture context. Not a git chat bot — structured and reviewable. likely_breakpoints are INVESTIGATIVE reasoning about what could break; validation_checks are what to verify after the change. Never remedial (no 'apply this fix'). release_note_draft is a draft the operator reviews.", changeBriefSchema.shape, (args) => wrap(handleChangeBrief(args, ctx)));
+    // PACK — ollama_incident_pack (deterministic orchestration, durable artifact)
+    server.tool("ollama_incident_pack", "PACK. Runs the full incident job end-to-end: triage_logs → corpus_search → incident_brief → deterministic markdown+JSON artifact on disk. Single call, single completed job, single pair of files the operator can keep and diff. Response is compact ({artifact:{markdown_path,json_path}, summary, steps}) — the full brief lives in the artifact, not the MCP payload. Fixed pipeline, fixed markdown layout, no prose drift. Use this instead of calling triage_logs + incident_brief manually when you want one deliverable.", incidentPackSchema.shape, (args) => wrap(handleIncidentPack(args, ctx)));
+    // PACK — ollama_repo_pack (onboarding job, corpus-first, durable artifact)
+    server.tool("ollama_repo_pack", "PACK. Runs the full repo ONBOARDING job: corpus_search (if corpus given) → repo_brief → targeted ollama_extract (narrow onboarding schema: packages, entrypoints, scripts, config_files, exposed_surfaces, runtime_hints) → deterministic markdown+JSON artifact on disk. Corpus-first posture: when a corpus is given, it's the main working surface alongside source_paths. Not repo Q&A — this is `get me onboarded fast with a stable operator artifact`. Response is compact; full brief + extracted facts live in the artifact.", repoPackSchema.shape, (args) => wrap(handleRepoPack(args, ctx)));
+    // PACK — ollama_change_pack (change-centered review job, durable artifact)
+    server.tool("ollama_change_pack", "PACK. Runs the full change REVIEW job: assemble evidence (diff + paths + corpus if given) → triage_logs ONLY when log_text is provided → change_brief → targeted ollama_extract (narrow review schema: scripts_touched, config_surfaces, runtime_hints) → deterministic markdown+JSON artifact on disk. Change-first, not repo-first — this is about the DELTA, not a tour. Release note draft is a blockquote-wrapped DRAFT (not marketing copy). No VCS integration — caller hands in diff_text / source_paths / optional log_text. Response is compact; full brief + extracted facts live in the artifact.", changePackSchema.shape, (args) => wrap(handleChangePack(args, ctx)));
+    // ARTIFACT — ollama_artifact_list (metadata-only index over pack artifacts)
+    server.tool("ollama_artifact_list", "ARTIFACT. Metadata-only index of pack artifacts on disk. Returns one compact record per artifact: `{pack, slug, title, created_at, weak, corpus_used, evidence_count, section_counts, md_path, json_path}`. Filter by pack / date_after / date_before / weak_only / strong_only; sort is newest first. Scans ~/.ollama-intern/artifacts/{incident,repo,change} by default; pass extra_artifact_dirs for additional read-only search surfaces. Full payloads belong to ollama_artifact_read — listing stays cheap.", artifactListSchema.shape, (args) => wrap(handleArtifactList(args, ctx)));
+    // ARTIFACT — ollama_artifact_read (typed read by identity or path)
+    server.tool("ollama_artifact_read", "ARTIFACT. Read a single pack artifact, typed by pack. Primary: `{pack, slug}` — identity-based, collisions fail loud. Secondary: `{json_path}` — absolute path, must live under a recognized artifact dir (canonical roots + extra_artifact_dirs), must end in .json, path-traversal rejected. Returns `{metadata, artifact}` where artifact is a discriminated union on `pack` (incident_pack / repo_pack / change_pack — payloads stay distinct, never flattened).", artifactReadSchema.shape, (args) => wrap(handleArtifactRead(args, ctx)));
+    // ARTIFACT — ollama_artifact_diff (structured same-pack comparison)
+    server.tool("ollama_artifact_diff", "ARTIFACT. Structured diff of two same-pack artifacts. Input: `{a: {pack, slug}, b: {pack, slug}}` — must share pack; cross-pack diffs refused loudly. Returns `{pack, a, b, weak, diff}` with weak flip surfaced at top level (strong→weak or weak→strong). Lists diff as {added, removed, unchanged} matched on primary key per item kind; narrative fields as {before, after}; release_note_draft also carries a compact LCS line_diff. Evidence is SUMMARIZED (counts + referenced_paths + path_delta), never exploded chunk-by-chunk. Deterministic ordering on every list.", artifactDiffSchema.shape, (args) => wrap(handleArtifactDiff(args, ctx)));
+    // ARTIFACT — ollama_artifact_export_to_path (handoff move, narrow writer)
+    server.tool("ollama_artifact_export_to_path", "ARTIFACT. Writes the artifact's EXISTING markdown to a caller-specified path with a provenance header prepended. No re-render, no model call. Path safety is strict: target_path must be absolute, must end in .md, must live under one of `allowed_roots` (REQUIRED — caller declares intent). Overwrite is opt-in: existing files refuse by default so re-runs never clobber hand-edits. Not a generic file writer — export is the single handoff move.", artifactExportToPathSchema.shape, (args) => wrap(handleArtifactExportToPath(args, ctx)));
+    // ARTIFACT — ollama_artifact_incident_note_snippet (operator note fragment)
+    server.tool("ollama_artifact_incident_note_snippet", "ARTIFACT. Renders a compact incident-note markdown fragment from an incident_pack artifact — top hypotheses, affected surfaces, next checks, with an evidence-aware operator tone. No model call, no re-render; pure derivation from stored JSON. Returns `{rendered, metadata}`. For the full artifact use artifact_read; for the whole markdown as a reviewable file use artifact_export_to_path.", artifactIncidentNoteSnippetSchema.shape, (args) => wrap(handleArtifactIncidentNoteSnippet(args, ctx)));
+    // ARTIFACT — ollama_artifact_onboarding_section_snippet (handbook fragment)
+    server.tool("ollama_artifact_onboarding_section_snippet", "ARTIFACT. Renders a handbook-ready `## What this repo is` section from a repo_pack artifact — thesis, key surfaces, read-next, runtime hints. Investigative tone preserved (read-next is LOOK AT, not prescriptive). No model call. Returns `{rendered, metadata}`.", artifactOnboardingSectionSnippetSchema.shape, (args) => wrap(handleArtifactOnboardingSectionSnippet(args, ctx)));
+    // ARTIFACT — ollama_artifact_release_note_snippet (change pack DRAFT fragment)
+    server.tool("ollama_artifact_release_note_snippet", "ARTIFACT. Renders the release-note draft from a change_pack artifact as a blockquote-wrapped DRAFT fragment with the caveat preserved. No model call, no polishing, no marketing lift. Returns `{rendered, metadata}` — operator reviews before publishing.", artifactReleaseNoteSnippetSchema.shape, (args) => wrap(handleArtifactReleaseNoteSnippet(args, ctx)));
+    // FLAGSHIP — ollama_embed_search (ephemeral concept search on ad-hoc candidates)
+    server.tool("ollama_embed_search", "FLAGSHIP. Rank AD-HOC candidates by concept similarity to a query. Pass `query` + `candidates: [{id, text}]`; server embeds everything, computes cosine, returns ranked `[{id, score, preview?}]`. Use this when you have in-memory candidates to compare; for persistent recall over memory/canon/doctrine use ollama_corpus_search instead. Does NOT return raw vectors to you.", embedSearchSchema.shape, (args) => wrap(handleEmbedSearch(args, ctx)));
+    // Corpus builder
+    server.tool("ollama_corpus_index", "Build a persistent named corpus. Pass `name` + `paths: string[]`; the server chunks, embeds, stores the corpus at ~/.ollama-intern/corpora/<name>.json AND writes a manifest at <name>.manifest.json that captures the declared paths + chunk params + embed model. Idempotent — unchanged files are reused by sha256; changed files are re-embedded; paths not in the input are dropped. For day-to-day upkeep once a manifest exists, prefer `ollama_corpus_refresh` — it reconciles corpus vs manifest and reports drift.", corpusIndexSchema.shape, (args) => wrap(handleCorpusIndex(args, ctx)));
+    // Corpus refresh — living-corpus workflow: reconcile against manifest
+    server.tool("ollama_corpus_refresh", "Reconcile a named corpus against its manifest (intent vs reality). Single arg: `name`. The manifest's declared paths, chunk params, and embed model are the source of truth — refresh doesn't accept them. Returns a drift report: added / changed / unchanged / deleted / missing (per-path lists) plus reused / reembedded / dropped (chunk-level counts) plus no_op. Idempotent: a no-change refresh is fast, boring, and makes zero embed calls.", corpusRefreshSchema.shape, (args) => wrap(handleCorpusRefresh(args, ctx)));
+    // Corpus list
+    server.tool("ollama_corpus_list", "List named corpora on disk with stats. No Ollama call. Use to discover what's been indexed, check freshness (indexed_at), or verify a corpus exists before searching.", corpusListSchema.shape, (args) => wrap(handleCorpusList(args, ctx)));
+    // LOW-LEVEL — ollama_embed (raw vectors for external index builds)
+    server.tool("ollama_embed", "LOW-LEVEL primitive. Returns raw 768-dim vectors for one text or a batch. Use `ollama_embed_search` instead for concept-search — this tool is for building external indexes (sqlite-vss, pgvector) where you need the raw geometry. Output can be large.", embedSchema.shape, (args) => wrap(handleEmbed(args, ctx)));
+    // Core — classify (batch-capable)
+    server.tool("ollama_classify", "Single-label classification with confidence. Single: pass `text`. BATCH: pass `items:[{id,text}]` — returns ONE envelope with `result.items[]` of `{id, ok, result|error}` plus `batch_count/ok_count/error_count`. Use the batch shape to chew through bulk labeling (commit types, PR titles, log severities) in one handoff instead of N round-trips. Set allow_none=true when weak guesses are worse than 'unsure' — label returns null below threshold (default 0.7).", classifySchema.shape, (args) => wrap(handleClassify(args, ctx)));
+    // Core — triage_logs (batch-capable)
+    server.tool("ollama_triage_logs", "Stable-shape log digest: {errors, warnings, suspected_root_cause}. Single: pass `log_text`. BATCH: pass `items:[{id,log_text}]` for triaging many log blobs at once (multiple CI runs, matrix legs, per-service logs) — returns one envelope with per-item entries. Use before grep-storms on long CI/test output.", triageLogsSchema.shape, (args) => wrap(handleTriageLogs(args, ctx)));
+    // Core — summarize_fast
+    server.tool("ollama_summarize_fast", "Gist of short input (best under ~4k tokens). Use as a decision gate: 'is this file worth reading in full?' Summary carries source_preview so you can spot-check fabrication.", summarizeFastSchema.shape, (args) => wrap(handleSummarizeFast(args, ctx)));
+    // Core — summarize_deep
+    server.tool("ollama_summarize_deep", "Digest of long input with optional focus. Pass EITHER `text` (when you already have the content) OR `source_paths[]` (server reads + chunks locally — use this to save Claude context). Exactly one of the two. Carries source_preview for fabrication spot-checks.", summarizeDeepSchema.shape, (args) => wrap(handleSummarizeDeep(args, ctx)));
+    // Core — draft
+    server.tool("ollama_draft", "DRAFT code or prose stubs (never autonomous — Claude reviews). Pass language for a server-side compile check: envelope returns {compiles, checker, stderr_tail}. target_path pointing into memory/, .claude/, docs/canon/, games/ requires confirm_write: true.", draftSchema.shape, (args) => wrap(handleDraft(args, ctx)));
+    // Core — extract (batch-capable)
+    server.tool("ollama_extract", "Schema-constrained JSON extraction using Ollama's JSON mode. Single: pass `text`, returns `{ok: true, data}` or `{ok: false, error: 'unparseable'}` — never partial. BATCH: pass `items:[{id,text}]` with a shared schema — returns one envelope with per-item `{id, ok, result|error}`. Use the batch shape for any 10+-similar-inputs workload (frontmatter, package.json, release metadata) so you hand over the whole job, not N calls.", extractSchema.shape, (args) => wrap(handleExtract(args, ctx)));
+    // Last resort — chat
+    server.tool("ollama_chat", "LAST RESORT catch-all. Prefer a specialty tool above when one fits. If you reach for this often, a specialty tool is missing and should be added.", chatSchema.shape, (args) => wrap(handleChat(args, ctx)));
+    return server;
+}
+async function main() {
+    const profile = loadProfile();
+    const ctx = {
+        client: new HttpOllamaClient(),
+        tiers: profile.tiers,
+        timeouts: profile.timeouts,
+        hardwareProfile: profile.name,
+        logger: new NdjsonLogger(),
+    };
+    // Profile-policy prewarm: pulls Instant tier into VRAM on dev profiles
+    // before connecting transport, so the first real Claude call doesn't eat
+    // cold-load latency. Failures are logged but never throw — server startup
+    // must not depend on Ollama being reachable.
+    if (profile.prewarm.length > 0) {
+        await runPrewarm(ctx, profile.prewarm);
+    }
+    const server = createServer(ctx);
+    const transport = new StdioServerTransport();
+    await server.connect(transport);
+}
+/**
+ * Run main() only when invoked as a script, not when imported by tests.
+ *
+ * Robust-on-Windows check: normalize both sides through realpathSync +
+ * fileURLToPath so forward/backslash and symlink differences don't cause
+ * the script to no-op silently (which is what Claude Code would see as
+ * "connected then immediately disconnected").
+ */
+function isInvokedAsScript() {
+    try {
+        if (!process.argv[1])
+            return false;
+        return realpathSync(fileURLToPath(import.meta.url)) === realpathSync(process.argv[1]);
+    }
+    catch {
+        return false;
+    }
+}
+if (isInvokedAsScript()) {
+    main().catch((err) => {
+        console.error(JSON.stringify(toErrorShape(err), null, 2));
+        process.exit(1);
+    });
+}
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AAEA;;;;;;GAMG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACjF,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAEvC,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AACvC,OAAO,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AAC5C,OAAO,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAC/C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC3C,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAG1C,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrE,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAC3E,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AACpF,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AACpF,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAClE,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AACjF,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AACjF,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AACpF,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAC3E,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AACpF,OAAO,EAAE,eAAe,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AACxE,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,+BAA+B,CAAC;AACvF,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3E,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,6BAA6B,CAAC;AACjF,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AACjF,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AACjF,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AACjF,OAAO,EAAE,0BAA0B,EAAE,0BAA0B,EAAE,MAAM,iCAAiC,CAAC;AACzG,OAAO,EACL,iCAAiC,EACjC,iCAAiC,EACjC,sCAAsC,EACtC,sCAAsC,EACtC,gCAAgC,EAChC,gCAAgC,GACjC,MAAM,6BAA6B,CAAC;AACrC,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAEzD,MAAM,UAAU,YAAY,CAAC,GAAe;IAC1C,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC,EAAE,IAAI,EAAE,mBAAmB,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,CAAC;IAE9E,MAAM,IAAI,GAAG,CAAI,CAAa,EAA+E,EAAE,CAC7G,CAAC,CAAC,IAAI,CACJ,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAC3F,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;QACR,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC;QACtF,OAAO,EAAE,IAAa;KACvB,CAAC,CACH,CAAC;IAEJ,6BAA6B;IAC7B,MAAM,CAAC,IAAI,CACT,iBAAiB,EACjB,+SAA+S,EAC/S,cAAc,CAAC,KAAK,EACpB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,cAAc,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC1C,CAAC;IAEF,iFAAiF;IACjF,MAAM,CAAC,IAAI,CACT,sBAAsB,EACtB,qbAAqb,EACrb,kBAAkB,CAAC,KAAK,EACxB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC9C,CAAC;IAEF,2EAA2E;IAC3E,MAAM,CAAC,IAAI,CACT,sBAAsB,EACtB,mlBAAmlB,EACnlB,kBAAkB,CAAC,KAAK,EACxB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC9C,CAAC;IAEF,+DAA+D;IAC/D,MAAM,CAAC,IAAI,CACT,uBAAuB,EACvB,+oBAA+oB,EAC/oB,mBAAmB,CAAC,KAAK,EACzB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,mBAAmB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC/C,CAAC;IAEF,wDAAwD;IACxD,MAAM,CAAC,IAAI,CACT,mBAAmB,EACnB,mlBAAmlB,EACnlB,eAAe,CAAC,KAAK,EACrB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,eAAe,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC3C,CAAC;IAEF,wEAAwE;IACxE,MAAM,CAAC,IAAI,CACT,qBAAqB,EACrB,inBAAinB,EACjnB,iBAAiB,CAAC,KAAK,EACvB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,iBAAiB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC7C,CAAC;IAEF,8EAA8E;IAC9E,MAAM,CAAC,IAAI,CACT,sBAAsB,EACtB,mgBAAmgB,EACngB,kBAAkB,CAAC,KAAK,EACxB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC9C,CAAC;IAEF,2EAA2E;IAC3E,MAAM,CAAC,IAAI,CACT,kBAAkB,EAClB,ugBAAugB,EACvgB,cAAc,CAAC,KAAK,EACpB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,cAAc,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC1C,CAAC;IAEF,2EAA2E;IAC3E,MAAM,CAAC,IAAI,CACT,oBAAoB,EACpB,+kBAA+kB,EAC/kB,gBAAgB,CAAC,KAAK,EACtB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC5C,CAAC;IAEF,4EAA4E;IAC5E,MAAM,CAAC,IAAI,CACT,sBAAsB,EACtB,mfAAmf,EACnf,kBAAkB,CAAC,KAAK,EACxB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC9C,CAAC;IAEF,mEAAmE;IACnE,MAAM,CAAC,IAAI,CACT,sBAAsB,EACtB,scAAsc,EACtc,kBAAkB,CAAC,KAAK,EACxB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC9C,CAAC;IAEF,oEAAoE;IACpE,MAAM,CAAC,IAAI,CACT,sBAAsB,EACtB,ijBAAijB,EACjjB,kBAAkB,CAAC,KAAK,EACxB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC9C,CAAC;IAEF,0EAA0E;IAC1E,MAAM,CAAC,IAAI,CACT,gCAAgC,EAChC,2bAA2b,EAC3b,0BAA0B,CAAC,KAAK,EAChC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,0BAA0B,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CACtD,CAAC;IAEF,4EAA4E;IAC5E,MAAM,CAAC,IAAI,CACT,uCAAuC,EACvC,qYAAqY,EACrY,iCAAiC,CAAC,KAAK,EACvC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,iCAAiC,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC7D,CAAC;IAEF,4EAA4E;IAC5E,MAAM,CAAC,IAAI,CACT,4CAA4C,EAC5C,qQAAqQ,EACrQ,sCAAsC,CAAC,KAAK,EAC5C,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,sCAAsC,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAClE,CAAC;IAEF,+EAA+E;IAC/E,MAAM,CAAC,IAAI,CACT,sCAAsC,EACtC,6PAA6P,EAC7P,gCAAgC,CAAC,KAAK,EACtC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,gCAAgC,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC5D,CAAC;IAEF,iFAAiF;IACjF,MAAM,CAAC,IAAI,CACT,qBAAqB,EACrB,mXAAmX,EACnX,iBAAiB,CAAC,KAAK,EACvB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,iBAAiB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC7C,CAAC;IAEF,iBAAiB;IACjB,MAAM,CAAC,IAAI,CACT,qBAAqB,EACrB,8fAA8f,EAC9f,iBAAiB,CAAC,KAAK,EACvB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,iBAAiB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC7C,CAAC;IAEF,sEAAsE;IACtE,MAAM,CAAC,IAAI,CACT,uBAAuB,EACvB,sbAAsb,EACtb,mBAAmB,CAAC,KAAK,EACzB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,mBAAmB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC/C,CAAC;IAEF,cAAc;IACd,MAAM,CAAC,IAAI,CACT,oBAAoB,EACpB,uKAAuK,EACvK,gBAAgB,CAAC,KAAK,EACtB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC5C,CAAC;IAEF,mEAAmE;IACnE,MAAM,CAAC,IAAI,CACT,cAAc,EACd,0PAA0P,EAC1P,WAAW,CAAC,KAAK,EACjB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CACvC,CAAC;IAEF,kCAAkC;IAClC,MAAM,CAAC,IAAI,CACT,iBAAiB,EACjB,4cAA4c,EAC5c,cAAc,CAAC,KAAK,EACpB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,cAAc,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC1C,CAAC;IAEF,qCAAqC;IACrC,MAAM,CAAC,IAAI,CACT,oBAAoB,EACpB,oTAAoT,EACpT,gBAAgB,CAAC,KAAK,EACtB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC5C,CAAC;IAEF,wBAAwB;IACxB,MAAM,CAAC,IAAI,CACT,uBAAuB,EACvB,8KAA8K,EAC9K,mBAAmB,CAAC,KAAK,EACzB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,mBAAmB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC/C,CAAC;IAEF,wBAAwB;IACxB,MAAM,CAAC,IAAI,CACT,uBAAuB,EACvB,qQAAqQ,EACrQ,mBAAmB,CAAC,KAAK,EACzB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,mBAAmB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAC/C,CAAC;IAEF,eAAe;IACf,MAAM,CAAC,IAAI,CACT,cAAc,EACd,iQAAiQ,EACjQ,WAAW,CAAC,KAAK,EACjB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CACvC,CAAC;IAEF,iCAAiC;IACjC,MAAM,CAAC,IAAI,CACT,gBAAgB,EAChB,6aAA6a,EAC7a,aAAa,CAAC,KAAK,EACnB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CACzC,CAAC;IAEF,qBAAqB;IACrB,MAAM,CAAC,IAAI,CACT,aAAa,EACb,mJAAmJ,EACnJ,UAAU,CAAC,KAAK,EAChB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CACtC,CAAC;IAEF,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,OAAO,GAAG,WAAW,EAAE,CAAC;IAC9B,MAAM,GAAG,GAAe;QACtB,MAAM,EAAE,IAAI,gBAAgB,EAAE;QAC9B,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,eAAe,EAAE,OAAO,CAAC,IAAI;QAC7B,MAAM,EAAE,IAAI,YAAY,EAAE;KAC3B,CAAC;IAEF,uEAAuE;IACvE,yEAAyE;IACzE,0EAA0E;IAC1E,6CAA6C;IAC7C,IAAI,OAAO,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC/B,MAAM,UAAU,CAAC,GAAG,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,MAAM,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC;IACjC,MAAM,SAAS,GAAG,IAAI,oBAAoB,EAAE,CAAC;IAC7C,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;AAClC,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,iBAAiB;IACxB,IAAI,CAAC;QACH,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;YAAE,OAAO,KAAK,CAAC;QACnC,OAAO,YAAY,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,YAAY,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IACxF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED,IAAI,iBAAiB,EAAE,EAAE,CAAC;IACxB,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QACnB,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QAC1D,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/observability.d.ts

@@ -0,0 +1,66 @@
+/**
+ * NDJSON observability — one line per call to ~/.ollama-intern/log.ndjson.
+ *
+ * This is what lets Claude tune delegation instead of guessing:
+ * "that call used 5k tokens on Deep when fast would have sufficed."
+ *
+ * Also logs timeout events and fallback decisions so later we can prove
+ * the system degraded correctly under pressure, not just that a call was slow.
+ */
+import type { Envelope, Residency } from "./envelope.js";
+import type { Tier } from "./tiers.js";
+export type LogEvent = {
+    kind: "call";
+    ts: string;
+    tool: string;
+    envelope: Envelope<unknown>;
+} | {
+    kind: "timeout";
+    ts: string;
+    tool: string;
+    tier: Tier;
+    timeout_ms: number;
+} | {
+    kind: "fallback";
+    ts: string;
+    tool: string;
+    from: Tier;
+    to: Tier;
+    reason: string;
+} | {
+    kind: "guardrail";
+    ts: string;
+    tool: string;
+    rule: string;
+    action: string;
+    detail?: unknown;
+} | {
+    kind: "prewarm";
+    ts: string;
+    tier: Tier;
+    model: string;
+    hardware_profile: string;
+    success: boolean;
+    elapsed_ms: number;
+    residency: Residency | null;
+    error?: string;
+};
+export interface Logger {
+    log(event: LogEvent): Promise<void>;
+}
+export declare class NdjsonLogger implements Logger {
+    private path;
+    private readyPromise;
+    constructor(path?: string);
+    private ready;
+    log(event: LogEvent): Promise<void>;
+}
+/** No-op logger for tests. */
+export declare class NullLogger implements Logger {
+    events: LogEvent[];
+    log(event: LogEvent): Promise<void>;
+}
+export declare function timestamp(): string;
+/** Build a LogEvent for a completed tool call from its envelope. */
+export declare function callEvent(tool: string, envelope: Envelope<unknown>): LogEvent;
+//# sourceMappingURL=observability.d.ts.map

package/dist/observability.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"observability.d.ts","sourceRoot":"","sources":["../src/observability.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAKH,OAAO,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AACzD,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAKvC,MAAM,MAAM,QAAQ,GAChB;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAA;CAAE,GACvE;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,IAAI,CAAC;IAAC,UAAU,EAAE,MAAM,CAAA;CAAE,GAC7E;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,IAAI,CAAC;IAAC,EAAE,EAAE,IAAI,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GACpF;IAAE,IAAI,EAAE,WAAW,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,OAAO,CAAA;CAAE,GAC/F;IACE,IAAI,EAAE,SAAS,CAAC;IAChB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,gBAAgB,EAAE,MAAM,CAAC;IACzB,OAAO,EAAE,OAAO,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,SAAS,GAAG,IAAI,CAAC;IAC5B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB,CAAC;AAEN,MAAM,WAAW,MAAM;IACrB,GAAG,CAAC,KAAK,EAAE,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACrC;AAED,qBAAa,YAAa,YAAW,MAAM;IAG7B,OAAO,CAAC,IAAI;IAFxB,OAAO,CAAC,YAAY,CAA8B;gBAE9B,IAAI,GAAE,MAAyB;IAEnD,OAAO,CAAC,KAAK;IAOP,GAAG,CAAC,KAAK,EAAE,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC;CAQ1C;AAED,8BAA8B;AAC9B,qBAAa,UAAW,YAAW,MAAM;IAChC,MAAM,EAAE,QAAQ,EAAE,CAAM;IACzB,GAAG,CAAC,KAAK,EAAE,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC;CAG1C;AAED,wBAAgB,SAAS,IAAI,MAAM,CAElC;AAED,oEAAoE;AACpE,wBAAgB,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,QAAQ,CAE7E"}

package/dist/observability.js

@@ -0,0 +1,51 @@
+/**
+ * NDJSON observability — one line per call to ~/.ollama-intern/log.ndjson.
+ *
+ * This is what lets Claude tune delegation instead of guessing:
+ * "that call used 5k tokens on Deep when fast would have sufficed."
+ *
+ * Also logs timeout events and fallback decisions so later we can prove
+ * the system degraded correctly under pressure, not just that a call was slow.
+ */
+import { appendFile, mkdir } from "node:fs/promises";
+import { homedir } from "node:os";
+import { dirname, join } from "node:path";
+const DEFAULT_LOG_DIR = join(homedir(), ".ollama-intern");
+const DEFAULT_LOG_PATH = process.env.INTERN_LOG_PATH || join(DEFAULT_LOG_DIR, "log.ndjson");
+export class NdjsonLogger {
+    path;
+    readyPromise = null;
+    constructor(path = DEFAULT_LOG_PATH) {
+        this.path = path;
+    }
+    ready() {
+        if (!this.readyPromise) {
+            this.readyPromise = mkdir(dirname(this.path), { recursive: true }).then(() => undefined);
+        }
+        return this.readyPromise;
+    }
+    async log(event) {
+        try {
+            await this.ready();
+            await appendFile(this.path, JSON.stringify(event) + "\n", "utf8");
+        }
+        catch {
+            // observability failures must never break tool calls
+        }
+    }
+}
+/** No-op logger for tests. */
+export class NullLogger {
+    events = [];
+    async log(event) {
+        this.events.push(event);
+    }
+}
+export function timestamp() {
+    return new Date().toISOString();
+}
+/** Build a LogEvent for a completed tool call from its envelope. */
+export function callEvent(tool, envelope) {
+    return { kind: "call", ts: timestamp(), tool, envelope };
+}
+//# sourceMappingURL=observability.js.map

package/dist/observability.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"observability.js","sourceRoot":"","sources":["../src/observability.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAI1C,MAAM,eAAe,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,gBAAgB,CAAC,CAAC;AAC1D,MAAM,gBAAgB,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,IAAI,CAAC,eAAe,EAAE,YAAY,CAAC,CAAC;AAuB5F,MAAM,OAAO,YAAY;IAGH;IAFZ,YAAY,GAAyB,IAAI,CAAC;IAElD,YAAoB,OAAe,gBAAgB;QAA/B,SAAI,GAAJ,IAAI,CAA2B;IAAG,CAAC;IAE/C,KAAK;QACX,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvB,IAAI,CAAC,YAAY,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC3F,CAAC;QACD,OAAO,IAAI,CAAC,YAAY,CAAC;IAC3B,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,KAAe;QACvB,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,UAAU,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,MAAM,CAAC,CAAC;QACpE,CAAC;QAAC,MAAM,CAAC;YACP,qDAAqD;QACvD,CAAC;IACH,CAAC;CACF;AAED,8BAA8B;AAC9B,MAAM,OAAO,UAAU;IACd,MAAM,GAAe,EAAE,CAAC;IAC/B,KAAK,CAAC,GAAG,CAAC,KAAe;QACvB,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC1B,CAAC;CACF;AAED,MAAM,UAAU,SAAS;IACvB,OAAO,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClC,CAAC;AAED,oEAAoE;AACpE,MAAM,UAAU,SAAS,CAAC,IAAY,EAAE,QAA2B;IACjE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;AAC3D,CAAC"}

package/dist/ollama.d.ts

@@ -0,0 +1,105 @@
+/**
+ * Ollama HTTP client — generate, embed, residency probe (/api/ps).
+ *
+ * Every call is guarded by the global semaphore. Timeouts via AbortController.
+ * Residency probe runs on every call so the envelope can surface eviction
+ * without the caller asking. This is what prevents silent 5–10× slowdowns
+ * from hiding behind polished output.
+ */
+import type { Residency } from "./envelope.js";
+export interface GenerateRequest {
+    model: string;
+    prompt: string;
+    system?: string;
+    format?: "json";
+    stream?: boolean;
+    options?: {
+        temperature?: number;
+        num_predict?: number;
+        top_p?: number;
+    };
+    /** Keep model resident for this long. "-1" = forever. */
+    keep_alive?: string | number;
+}
+export interface GenerateResponse {
+    model: string;
+    response: string;
+    done: boolean;
+    prompt_eval_count?: number;
+    eval_count?: number;
+    total_duration?: number;
+    load_duration?: number;
+    eval_duration?: number;
+}
+export interface ChatMessage {
+    role: "system" | "user" | "assistant";
+    content: string;
+}
+export interface ChatRequest {
+    model: string;
+    messages: ChatMessage[];
+    format?: "json";
+    stream?: boolean;
+    options?: GenerateRequest["options"];
+    keep_alive?: string | number;
+}
+export interface ChatResponse {
+    model: string;
+    message: ChatMessage;
+    done: boolean;
+    prompt_eval_count?: number;
+    eval_count?: number;
+}
+export interface EmbedRequest {
+    model: string;
+    input: string | string[];
+    keep_alive?: string | number;
+}
+export interface EmbedResponse {
+    model: string;
+    embeddings: number[][];
+}
+export interface PsModel {
+    name: string;
+    model: string;
+    size: number;
+    size_vram: number;
+    digest: string;
+    expires_at: string;
+}
+export interface PsResponse {
+    models: PsModel[];
+}
+export interface OllamaClient {
+    generate(req: GenerateRequest, signal?: AbortSignal): Promise<GenerateResponse>;
+    chat(req: ChatRequest, signal?: AbortSignal): Promise<ChatResponse>;
+    embed(req: EmbedRequest, signal?: AbortSignal): Promise<EmbedResponse>;
+    residency(model: string): Promise<Residency | null>;
+}
+/**
+ * Normalize an OLLAMA_HOST value.
+ *
+ * Ollama's CLI accepts `127.0.0.1:11434` (no scheme) and routes Windows users
+ * often set it that way. fetch() requires a full URL, so we coerce anything
+ * that looks like host[:port] into a proper http:// URL, and strip trailing
+ * slashes so path concatenation stays clean.
+ */
+export declare function normalizeOllamaHost(raw: string | undefined): string;
+export declare class HttpOllamaClient implements OllamaClient {
+    private baseUrl;
+    constructor(baseUrl?: string);
+    generate(req: GenerateRequest, signal?: AbortSignal): Promise<GenerateResponse>;
+    chat(req: ChatRequest, signal?: AbortSignal): Promise<ChatResponse>;
+    embed(req: EmbedRequest, signal?: AbortSignal): Promise<EmbedResponse>;
+    residency(model: string): Promise<Residency | null>;
+    private post;
+    private get;
+}
+export declare function countTokens(resp: {
+    prompt_eval_count?: number;
+    eval_count?: number;
+}): {
+    in: number;
+    out: number;
+};
+//# sourceMappingURL=ollama.d.ts.map

package/dist/ollama.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ollama.d.ts","sourceRoot":"","sources":["../src/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAIH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AAI/C,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,OAAO,CAAC,EAAE;QACR,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,CAAC;IACF,yDAAyD;IACzD,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,OAAO,CAAC;IACd,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;IACtC,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,OAAO,CAAC,EAAE,eAAe,CAAC,SAAS,CAAC,CAAC;IACrC,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,WAAW,CAAC;IACrB,IAAI,EAAE,OAAO,CAAC;IACd,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,EAAE,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,OAAO,EAAE,CAAC;CACnB;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,GAAG,EAAE,eAAe,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAChF,IAAI,CAAC,GAAG,EAAE,WAAW,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IACpE,KAAK,CAAC,GAAG,EAAE,YAAY,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,aAAa,CAAC,CAAC;IACvE,SAAS,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC,CAAC;CACrD;AAED;;;;;;;GAOG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,CAMnE;AAED,qBAAa,gBAAiB,YAAW,YAAY;IACnD,OAAO,CAAC,OAAO,CAAS;gBACZ,OAAO,CAAC,EAAE,MAAM;IAItB,QAAQ,CAAC,GAAG,EAAE,eAAe,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAI/E,IAAI,CAAC,GAAG,EAAE,WAAW,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC;IAInE,KAAK,CAAC,GAAG,EAAE,YAAY,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,aAAa,CAAC;IAItE,SAAS,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC;YAkB3C,IAAI;YA2CJ,GAAG;CAalB;AAED,wBAAgB,WAAW,CAAC,IAAI,EAAE;IAAE,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG;IACtF,EAAE,EAAE,MAAM,CAAC;IACX,GAAG,EAAE,MAAM,CAAC;CACb,CAEA"}

package/dist/ollama.js

@@ -0,0 +1,109 @@
+/**
+ * Ollama HTTP client — generate, embed, residency probe (/api/ps).
+ *
+ * Every call is guarded by the global semaphore. Timeouts via AbortController.
+ * Residency probe runs on every call so the envelope can surface eviction
+ * without the caller asking. This is what prevents silent 5–10× slowdowns
+ * from hiding behind polished output.
+ */
+import { ollamaSemaphore } from "./semaphore.js";
+import { InternError } from "./errors.js";
+const API_TIMEOUT_MS = 10_000;
+/**
+ * Normalize an OLLAMA_HOST value.
+ *
+ * Ollama's CLI accepts `127.0.0.1:11434` (no scheme) and routes Windows users
+ * often set it that way. fetch() requires a full URL, so we coerce anything
+ * that looks like host[:port] into a proper http:// URL, and strip trailing
+ * slashes so path concatenation stays clean.
+ */
+export function normalizeOllamaHost(raw) {
+    const fallback = "http://127.0.0.1:11434";
+    const value = (raw ?? "").trim();
+    if (!value)
+        return fallback;
+    const withScheme = /^https?:\/\//i.test(value) ? value : `http://${value}`;
+    return withScheme.replace(/\/+$/, "");
+}
+export class HttpOllamaClient {
+    baseUrl;
+    constructor(baseUrl) {
+        this.baseUrl = normalizeOllamaHost(baseUrl ?? process.env.OLLAMA_HOST);
+    }
+    async generate(req, signal) {
+        return this.post("/api/generate", { ...req, stream: false }, signal);
+    }
+    async chat(req, signal) {
+        return this.post("/api/chat", { ...req, stream: false }, signal);
+    }
+    async embed(req, signal) {
+        return this.post("/api/embed", req, signal);
+    }
+    async residency(model) {
+        try {
+            const ps = await this.get("/api/ps");
+            const hit = ps.models.find((m) => m.name === model || m.model === model);
+            if (!hit)
+                return null;
+            const evicted = hit.size_vram < hit.size;
+            return {
+                in_vram: !evicted,
+                size_bytes: hit.size,
+                size_vram_bytes: hit.size_vram,
+                evicted,
+                expires_at: hit.expires_at ?? null,
+            };
+        }
+        catch {
+            return null;
+        }
+    }
+    async post(path, body, signal) {
+        const release = await ollamaSemaphore.acquire();
+        try {
+            const res = await fetch(`${this.baseUrl}${path}`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify(body),
+                signal,
+            });
+            if (!res.ok) {
+                const text = await res.text().catch(() => "");
+                if (res.status === 404) {
+                    throw new InternError("OLLAMA_MODEL_MISSING", `Model not found (404): ${text}`, "Run `ollama pull <model>` — see README for the full tier list.", false);
+                }
+                throw new InternError("OLLAMA_UNREACHABLE", `Ollama returned ${res.status}: ${text}`, "Check that Ollama is running (`ollama serve`) and reachable at OLLAMA_HOST.", true);
+            }
+            return (await res.json());
+        }
+        catch (err) {
+            if (err instanceof InternError)
+                throw err;
+            if (err.name === "AbortError") {
+                throw new InternError("OLLAMA_TIMEOUT", "Ollama request aborted", "Tier timeout fired — see fallback in envelope.", true);
+            }
+            throw new InternError("OLLAMA_UNREACHABLE", `Failed to reach Ollama: ${err.message}`, "Check that Ollama is running at OLLAMA_HOST (default http://127.0.0.1:11434).", true);
+        }
+        finally {
+            release();
+        }
+    }
+    async get(path) {
+        const controller = new AbortController();
+        const timeout = setTimeout(() => controller.abort(), API_TIMEOUT_MS);
+        try {
+            const res = await fetch(`${this.baseUrl}${path}`, { signal: controller.signal });
+            if (!res.ok) {
+                throw new InternError("OLLAMA_UNREACHABLE", `Ollama GET ${path} → ${res.status}`, "Check Ollama is running.", true);
+            }
+            return (await res.json());
+        }
+        finally {
+            clearTimeout(timeout);
+        }
+    }
+}
+export function countTokens(resp) {
+    return { in: resp.prompt_eval_count ?? 0, out: resp.eval_count ?? 0 };
+}
+//# sourceMappingURL=ollama.js.map

package/dist/ollama.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ollama.js","sourceRoot":"","sources":["../src/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAG1C,MAAM,cAAc,GAAG,MAAM,CAAC;AAiF9B;;;;;;;GAOG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAuB;IACzD,MAAM,QAAQ,GAAG,wBAAwB,CAAC;IAC1C,MAAM,KAAK,GAAG,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACjC,IAAI,CAAC,KAAK;QAAE,OAAO,QAAQ,CAAC;IAC5B,MAAM,UAAU,GAAG,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,EAAE,CAAC;IAC3E,OAAO,UAAU,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AACxC,CAAC;AAED,MAAM,OAAO,gBAAgB;IACnB,OAAO,CAAS;IACxB,YAAY,OAAgB;QAC1B,IAAI,CAAC,OAAO,GAAG,mBAAmB,CAAC,OAAO,IAAI,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC;IACzE,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,GAAoB,EAAE,MAAoB;QACvD,OAAO,IAAI,CAAC,IAAI,CAAoC,eAAe,EAAE,EAAE,GAAG,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,MAAM,CAAC,CAAC;IAC1G,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,GAAgB,EAAE,MAAoB;QAC/C,OAAO,IAAI,CAAC,IAAI,CAA4B,WAAW,EAAE,EAAE,GAAG,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,MAAM,CAAC,CAAC;IAC9F,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,GAAiB,EAAE,MAAoB;QACjD,OAAO,IAAI,CAAC,IAAI,CAA8B,YAAY,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;IAC3E,CAAC;IAED,KAAK,CAAC,SAAS,CAAC,KAAa;QAC3B,IAAI,CAAC;YACH,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC,GAAG,CAAa,SAAS,CAAC,CAAC;YACjD,MAAM,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,KAAK,IAAI,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,CAAC;YACzE,IAAI,CAAC,GAAG;gBAAE,OAAO,IAAI,CAAC;YACtB,MAAM,OAAO,GAAG,GAAG,CAAC,SAAS,GAAG,GAAG,CAAC,IAAI,CAAC;YACzC,OAAO;gBACL,OAAO,EAAE,CAAC,OAAO;gBACjB,UAAU,EAAE,GAAG,CAAC,IAAI;gBACpB,eAAe,EAAE,GAAG,CAAC,SAAS;gBAC9B,OAAO;gBACP,UAAU,EAAE,GAAG,CAAC,UAAU,IAAI,IAAI;aACnC,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,IAAI,CAAC;QACd,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,IAAI,CAAa,IAAY,EAAE,IAAU,EAAE,MAAoB;QAC3E,MAAM,OAAO,GAAG,MAAM,eAAe,CAAC,OAAO,EAAE,CAAC;QAChD,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE;gBAChD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;gBAC1B,MAAM;aACP,CAAC,CAAC;YACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;gBACZ,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC;gBAC9C,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;oBACvB,MAAM,IAAI,WAAW,CACnB,sBAAsB,EACtB,0BAA0B,IAAI,EAAE,EAChC,gEAAgE,EAChE,KAAK,CACN,CAAC;gBACJ,CAAC;gBACD,MAAM,IAAI,WAAW,CACnB,oBAAoB,EACpB,mBAAmB,GAAG,CAAC,MAAM,KAAK,IAAI,EAAE,EACxC,6EAA6E,EAC7E,IAAI,CACL,CAAC;YACJ,CAAC;YACD,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAS,CAAC;QACpC,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAI,GAAG,YAAY,WAAW;gBAAE,MAAM,GAAG,CAAC;YAC1C,IAAK,GAAa,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;gBACzC,MAAM,IAAI,WAAW,CAAC,gBAAgB,EAAE,wBAAwB,EAAE,gDAAgD,EAAE,IAAI,CAAC,CAAC;YAC5H,CAAC;YACD,MAAM,IAAI,WAAW,CACnB,oBAAoB,EACpB,2BAA4B,GAAa,CAAC,OAAO,EAAE,EACnD,+EAA+E,EAC/E,IAAI,CACL,CAAC;QACJ,CAAC;gBAAS,CAAC;YACT,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,GAAG,CAAI,IAAY;QAC/B,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,cAAc,CAAC,CAAC;QACrE,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;YACjF,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;gBACZ,MAAM,IAAI,WAAW,CAAC,oBAAoB,EAAE,cAAc,IAAI,MAAM,GAAG,CAAC,MAAM,EAAE,EAAE,0BAA0B,EAAE,IAAI,CAAC,CAAC;YACtH,CAAC;YACD,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAM,CAAC;QACjC,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,OAAO,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;CACF;AAED,MAAM,UAAU,WAAW,CAAC,IAAyD;IAInF,OAAO,EAAE,EAAE,EAAE,IAAI,CAAC,iBAAiB,IAAI,CAAC,EAAE,GAAG,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC,EAAE,CAAC;AACxE,CAAC"}

package/dist/prewarm.d.ts

@@ -0,0 +1,28 @@
+/**
+ * Targeted prewarm — pulls profile.prewarm tiers into VRAM at server startup
+ * so the first real Claude call doesn't eat cold-load latency.
+ *
+ * Behavior is profile policy, not magic:
+ *   - dev-rtx5080:        prewarm = ["instant"]
+ *   - dev-rtx5080-llama:  prewarm = ["instant"]
+ *   - m5-max:             prewarm = []  (cold-load on unified memory is ~free)
+ *
+ * Each prewarm attempt logs a {kind: "prewarm", ...} NDJSON event with
+ * model, success/failure, elapsed_ms, and residency. That keeps benchmarks
+ * able to distinguish cold from warm and keeps adoption data clean.
+ *
+ * Failures are logged but never thrown — server startup must not depend on
+ * Ollama being reachable. If the model is missing or Ollama is down, the
+ * server still comes up; the user sees the prewarm failure in the log.
+ */
+import { type Tier } from "./tiers.js";
+import type { RunContext } from "./runContext.js";
+/**
+ * Run prewarm for the given tiers. Each tier issues a minimal generate
+ * (`prompt: "ok", num_predict: 1`) with `keep_alive: -1` so the model
+ * stays resident afterwards.
+ *
+ * Returns the count of successful prewarms; never throws.
+ */
+export declare function runPrewarm(ctx: RunContext, tiers: Tier[]): Promise<number>;
+//# sourceMappingURL=prewarm.d.ts.map

package/dist/prewarm.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"prewarm.d.ts","sourceRoot":"","sources":["../src/prewarm.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAe,KAAK,IAAI,EAAE,MAAM,YAAY,CAAC;AACpD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAMlD;;;;;;GAMG;AACH,wBAAsB,UAAU,CAAC,GAAG,EAAE,UAAU,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CA0ChF"}