Skip to content

Releases: OutVersus/memoryrail

Release list

v0.1.1

v0.1.1 Latest
Latest

Choose a tag to compare

View all tags
@OutVersus OutVersus released this 10 Oct 04:51
v0.1.1
602f1e4

Adds setup docs for OpenAI Codex and GitHub Copilot, plus two small features found while writing them.

Update: npm install -g memoryrail@latest

New

  • Setup guide for Codex and Copilot (VS Code, Copilot CLI, Copilot cloud agent): docs/CLIENTS.md. Written from each vendor's documentation, which the guide links.
  • memoryrail sync --also <path> writes the managed block into extra instruction files, for example .github/copilot-instructions.md. Paths outside the repository are refused.
  • memoryrail doctor warns when AGENTS.md is larger than 32 KiB, the default amount Codex reads before it stops.

Known limitations

  • The Codex and Copilot configs have not been run against those products by us. If one doesn't work, please open an issue with the client version and the error.
  • Search is still keyword-based, and the secret scanner is a pattern check, not a guarantee.
  • memoryrail install still configures Claude Code and Cursor only.

Full changes: v0.1.0...v0.1.1

Assets 2
Loading

v0.1.0

v0.1.0

Choose a tag to compare

View all tags
@OutVersus OutVersus released this 10 Oct 02:24
v0.1.0
531ec0e

First public release of MemoryRail: git-native project memory for coding agents.

Install: npm install -g memoryrail (Node 20+). Docs: https://memoryrail.si

What's in it

  • Memories as plain Markdown in .memoryrail/: decision, constraint, gotcha, attempt (failed approach), thread, session
  • precheck: before changing code, surfaces the rules, failed attempts, gotchas and decisions that apply to a plan or files (--staged --fail for git hooks)
  • MCP server (memoryrail serve) with 8 tools, plus a CLI with the same commands
  • sync writes AGENTS.md, CLAUDE.md and Cursor rules from memory; sync --check for CI
  • Optional review gate for agent-written memories; secret scanning on write; lint and doctor
  • Published format spec and JSON Schema

Known limitations

  • Early (0.x): the format and commands may change before 1.0
  • Search is keyword-based (BM25), with no synonym or semantic matching
  • The secret scanner is a pattern check, not a guarantee
  • memoryrail install configures Claude Code and Cursor only
  • Little real-world use so far

Ideas (not code) borrowed from projectmem, agent-memory and Memory Trail; see the README credits.

Loading