Skip to content

v0.1.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 07 Sep 08:48
· 19 commits to master since this release
v0.1.0
a591dbe

CogZ v0.1.0

CogZ is a local-first, code-aware engineering cognition runtime for AI coding agents. It gives a coding agent persistent memory, contextual retrieval, and continuous cognition about a software repository — all running locally, no cloud services required.

Features

  • Persistent memory — observations, rules, and knowledge stored as Markdown files, structured by taxonomy and linked to the code itself
  • Code-aware search — hybrid FTS5 + vector search with graph expansion, separate embedding models for code and knowledge
  • Context packs — scoped, ranked, token-budgeted context with graph provenance, assembled in three modes (cold start, task, escalation)
  • Code indexing — tree-sitter parses Rust, Python, Go, JavaScript, TypeScript, TSX, and Bash into a graph of functions, classes, files, and modules with structural edges
  • Consolidation — dedup detection, NLI-based contradiction flagging, observation-to-rule promotion, duplicate merging
  • Lifecycle hooks — session start, prompt submit, file save, session end — injects context packs and triggers incremental reindexing
  • Graceful degradation — works fully without ONNX models in FTS-only mode; models auto-download on first use and auto-unload when idle
  • Agent-agnostic — MCP server (13 tools) + shell hooks, works with Devin, Claude Code, Cursor, Codex, Windsurf, and any MCP-compatible agent

Install

curl -fsSL https://raw.githubusercontent.com/balaianu/CogZ/master/install.sh | bash

Quick start

cd ~/your-project
cogz init
cogz index
cogz search "authentication flow"

Requirements

  • Linux x86_64/aarch64, macOS arm64, or Windows x86_64
  • ~550 MB disk (binary + ONNX Runtime + 3 models)
  • ~11 MB RAM idle, ~300-500 MB with models loaded

Documentation

Checksums

See SHA256SUMS in the release assets for binary integrity verification.

Assets 11
Loading