v1.0.0

RepoMind v1.0.0

Free-path code intelligence season close.

Included

• Program eval n=52 with difficulty predicates (mini / mini_js / production_rag)
• Language packs: python-ast, js, json (stdlib; no grammar download in default CI)
• Rename-aware history via committed renames.jsonl
• Offline repomind pack / unpack (billed $0)
• Catalog UI: indexer_version, tree_hash, source_sha
• production_rag pin: bf6e36d1d4ca353c4f17f649cb721da51d74f6bb
• Load: docs/assets/load.json (200 mini asks, p50/p95; lexical fixture honesty)

Still PLANNED

• No live git on host (snapshots only)
• No stranger uploads / raw paths
• No required tree-sitter in CI
• Not semantic/SOTA code search

Host: https://pax-repomind.vercel.app

v0.3.0

RepoMind v0.3.0

• History without git(1): committed fixture snapshots → hosted GET /v1/code/history returns 200 (ADR 0004)
• Pinned P1 dogfood: production_rag from d43f812…; catalog source_sha
• Who calls X: Python AST incoming refs + UI /refs
• Goldens 14/14 + 8/8 + JS, billed $0
• Hosted: https://pax-repomind.vercel.app

v0.2.0

Catalog contract (production_rag on POST), content-addressed incremental index, GET /v1/catalog, free-path mini_js, optional git history with capability_missing. Goldens 14/14 + 8/8 + JS, billed_usd 0. Hosted: https://pax-repomind.vercel.app

RepoMind v0.1.0

Offline Python code navigation with AST-aware chunks, a closed read-only catalog, path:line citations, deterministic refusal, CLI/API/UI, 14/14 mini evals, and 8/8 Production RAG snapshot dogfood evals. No embeddings, hosted LLM, provider keys, model judge, or billed calls. The dogfood score is a locate-and-cite regression gate on committed snapshot bytes, not a claim about arbitrary-repository retrieval quality.