v1.89.14

Release v1.89.14

v1.89.13

fix(distill): emit hub-publishable capsule fields for conversation_distillation capsules (pairs with evomap-hub #1461 schema/antiCheat relaxation that accepts diff-less conversation capsules)

v1.89.11

Release v1.89.11

v1.89.10

@evomap/evolver v1.89.10

Format-aware proxy routing now covers all major coding-agent API surfaces — each dialect forwards to its native upstream, no translation.

Added (#254)

• GET /v1/models probe → Anthropic or OpenAI model list (by anthropic-version header); codex/opencode/cursor/SDK startup probes no longer 404.
• Ollama → POST /api/chat + /api/generate (EVOMAP_OLLAMA_BASE_URL, default 127.0.0.1:11434), NDJSON streaming.
• Vertex AI Gemini → native /v1/projects/.../locations/.../models:… path, region base, OAuth Bearer (EVOMAP_VERTEX_ACCESS_TOKEN).
• opencode / Kiro trace client detection (Kiro = managed AWS backend → captured via session hooks, not the proxy).

New routes are inert until their upstream key is configured — no impact on existing traffic.

Upgrade: npm i -g @evomap/evolver@latest

v1.89.9

@evomap/evolver v1.89.9

Format-aware proxy routing now covers four dialects — each goes to its native upstream, no translation.

Added

• Gemini (#252): native path POST /v1beta/models/<model>:generateContent / :streamGenerateContent → Google Generative Language API (x-goog-api-key via EVOMAP_GEMINI_API_KEY/GEMINI_API_KEY/GOOGLE_API_KEY). Point the Gemini CLI / google-genai SDK base URL at the proxy.
• OpenAI Chat Completions (#253): POST /v1/chat/completions → OpenAI /chat/completions (cursor's OpenAI mode + generic OpenAI Chat clients).

All four routes (Anthropic /v1/messages, OpenAI Responses /v1/responses, OpenAI Chat /v1/chat/completions, Gemini /v1beta/models/...) capture usage / finish reason / threading / streaming. New routes 401 until their upstream key is set — no impact on existing traffic.

Upgrade: npm i -g @evomap/evolver@latest

v1.89.8

@evomap/evolver v1.89.8 — hotfix (completes 1.89.7)

Fixes the startup crash that affected both v1.89.6 and v1.89.7.

Fixed

• evolver run crashed at require — savingsCore.js shipped without the conformance/savings-core/constants.json it loads. It had to be added to TWO allow-lists: public.manifest.json (copy into the build) and the published package.json files field (what npm publish ships). v1.89.7 fixed only the first and still shipped broken; v1.89.8 fixes both.
• built-entry-loads pre-publish gate hardened — it now npm packs + extracts the tarball and requires the entry from the EXTRACTED package (deps via NODE_PATH), so a files-field exclusion blocks the release. Verified against a fresh npm install.

Upgrade: npm i -g @evomap/evolver@latest

v1.89.7

@evomap/evolver v1.89.7 — hotfix

Fixes a startup crash in v1.89.6.

Fixed

• evolver run crashed at require on v1.89.6 — the package shipped savingsCore.js but the publish manifest omitted conformance/, so the required conformance/savings-core/constants.json was missing from the tarball and index.js threw at load. conformance/** is now packaged.
• New built-entry-loads pre-publish gate requires the built dist-public/index.js so a missing packaged runtime file blocks the release instead of shipping (parse-only checks + source-tree tests missed it).

Upgrade: npm i -g @evomap/evolver@latest.

v1.89.6

@evomap/evolver v1.89.6

Proxy trace data-quality fixes so Claude Code, cursor, and codex traces are actually usable, plus hub-decryptability fail-closed (#251).

Fixed

• Valid-JSON truncation — over-cap bodies are wrapped in a parseable envelope instead of hard-cut; field cap 64KB→1MB, upload 256KB→4MB.
• Session threading — per-session session_id extracted from metadata.user_id / x-cursor-session-id; rows now carry a trace id.
• Streaming usage — end-of-stream token usage / finish reason / provider response id captured by a passive SSE tee (bytes forwarded unchanged), on both /v1/messages and /v1/responses. Codex sessions thread by previous_response_id → response.id.
• OpenAI/codex shapes — prompt_tokens/completion_tokens and Responses status/incomplete_details.reason recognized.
• Hub-decryptable uploads (fail-closed) — when profile analysis is enabled, only rows carrying a hub-decryptable hub_key_envelope reach the warehouse; the hub trace public key can be distributed via a signature-verified config. No effect when analysis is disabled (the default).

v1.89.5

Release v1.89.5

v1.89.3

Release v1.89.3