engram v0.2.0-beta.2

engram v0.2.0-beta.2 Release Notes

v0.2.0-beta.2 is the first beta candidate focused on installability and first-run onboarding after
the v0.2.0-beta.1 review.

What Changed

• Added engram setup, a guided dry-run-first setup path for Claude Code, Codex, and Cursor.
• Added first-orient knowledge onboarding: when no documents are indexed yet, the agent is told to
  ask which existing docs, runbooks, notes, or knowledge folders should be ingested.
• Added engram warmup embeddings for explicit local embedding-cache preparation before first use
  or offline work.
• Improved cold embedding-cache warnings, RocksDB lock recovery messages, and daemon startup
  failure diagnostics.
• Prepared Homebrew packaging for macOS Apple Silicon beta installs.
• Expanded README, MCP setup, security, and contribution guidance for an open-source beta.

Install

Homebrew is the preferred macOS Apple Silicon path once the tap is published:

brew tap ymeiri/engram
brew install engram
engram init
engram setup

Release tarballs remain available from GitHub Releases for users who do not use Homebrew.

First Run

1. Run engram warmup embeddings if you want to prepare the local embedding model cache up front.
2. Run engram init.
3. Run engram setup --agent <claude-code|codex|cursor> and review the dry-run.
4. Re-run setup with --write after approving the planned files.
5. Restart the agent and ask it to run orient.
6. If no documents are indexed yet, the orient packet asks the agent to collect approved knowledge
   paths from the user before ingestion.

Known Limitations

• Homebrew packaging is scoped to macOS Apple Silicon for this beta.
• Setup writes are approval-gated. Dry-run is the default.
• Engram stores data locally under ~/.engram/ unless a project-specific or explicit data path is
  used.
• The first embedding use may download all-MiniLM-L6-v2 from Hugging Face unless the cache has
  already been warmed.