Summary

PlanWeave is a file-backed loop engineering system for long-running coding agents. It transforms fuzzy plans into a structured task graph of claimable blocks, routing them through implementation and review agents within a recoverable local loop. This system enhances complex engineering workflows by providing durable run artifacts, parallel execution, staged checks, and iterative feedback.

How It Works

PlanWeave models projects as a task graph where files are nodes and documents are actionable blocks. This graph-centric approach enables visible, editable task flows, dependencies, and review loops. Agents claim individual blocks for focused implementation or review, receiving scoped graph context. Durable run artifacts are recorded locally, ensuring a recoverable history. This design facilitates parallel implementation, staged checks, and iterative feedback within a single, local environment.

Quick Start & Requirements

• Installation: CLI via npm install -g @planweave-ai/cli or Homebrew brew install GaosCode/tap/planweave. Agent skills: npx skills@latest add GaosCode/PlanWeave.
• Prerequisites: Node.js, Homebrew, agent authentication (e.g., OpenAI API keys).
• Desktop App: Available from GitHub Releases; installers are unsigned and may require manual trust/bypasses (e.g., xattr -dr com.apple.quarantine on macOS).
• Docs: DEVELOPMENT.md for source setup.

Highlighted Details

• File-backed Graph Model: Projects are fundamentally task graphs, not chat overlays.
• Zero-Config Start: Enables immediate use with CLI and agent skills in existing projects.
• Scoped Context: Agents receive only the current block and relevant graph context for focus.
• Flexible Agent Routing: Supports diverse executors (Codex, Claude Code, Pi, Grok, local scripts) per block.
• MCP & ChatGPT Integration: Local MCP server allows ChatGPT to author, inspect, and manage plans.
• Full Auto-Run: Capable of autonomous execution, review, feedback handling, and recovery.
• First-Class Review: Structured feedback loops integrate directly into the workflow.
• Local-First: Plans, prompts, and run records remain inspectable locally.
• Live Observability: Detailed event logs, statistics, search, and todo views are available.

Maintenance & Community

The README does not detail community channels, active contributors, sponsorships, or a public roadmap. Development details are referenced in DEVELOPMENT.md.

Licensing & Compatibility

• License: MIT.
• Compatibility: Permissive MIT license allows for commercial use and integration with closed-source projects.

Limitations & Caveats

Desktop application installers are unsigned, requiring manual trust and potential security bypasses. Agent execution relies on external authentication, handled by the provider. Headless MCP server setup requires systemd configuration.