Skip to content

Releases: geronimo-iia/claude-local-stack

Release list

v1.0.0 — First stable release

v1.0.0 — First stable release Latest
Latest

Choose a tag to compare

View all tags
@geronimo-iia geronimo-iia released this 06 Sep 06:52
v1.0.0
39816fb

First stable, fully documented release. Marks the completion of the bifrost migration and the architectural foundation for multi-profile local+cloud AI routing on Apple Silicon.

What's in this release

Core architecture

  • Bifrost replaces LiteLLM as the routing gateway across all profiles — correctly forwards tools to OpenAI-compatible backends (LiteLLM silently stripped them)
  • Per-service env layering — ai-stack.env (topology) → lib/services/<svc>/default.env → config/profiles/<profile>/<svc>.env; AWS credentials scoped to the services that need them
  • Profile-driven install — ai-install is bootstrap-only; ai-stack install reads services.yaml per active profile
  • config/.env local override — gitignored, sourced last, for machine-specific credentials and ports
  • supervised-launch — bounded crash recovery (5 restarts / 60s window) across all services; overmind auto-restart deliberately disabled

Profiles

  • local — offline, Qwen3.6-35B-A3B-4bit via rapid-mlx with MTP (~73 t/s)
  • max / max-direct — Anthropic API via Claude Max plan
  • bedrock / bedrock-direct — AWS Bedrock with SigV4 auth
  • mistral / mistral-light — Ollama backends
  • rapid-mlx-test — model benchmarking profile

Docs

  • Architecture decision records (docs/decisions/) — 5 ADRs documenting why each key choice was made
  • docs/invariants.md — rules that must hold to avoid silent failures
  • docs/troubleshooting.md — common failure scenarios with diagnostic commands
  • AGENTS.md — full stack reference for AI agents working in this repo
  • Cross-linked docs throughout

Requirements

  • Apple Silicon (M1/M2/M3/M4), macOS
  • asdf, overmind, SOPS
  • Node.js (for bifrost via npx), Python (for rapid-mlx), Go toolchain optional
Assets 2
Loading