pi-fabric plugin
An extension for the Pi agent: one programmable fabric_exec tool calls tools, MCP, agents and workflows from code
- 304 ★ on the repository
- Last commit
Broad access
Why this level
- Agent programs run as the session and call Pi tools, including file writes and commands
- Choosing CPython or the Node or Bun modes removes the sandbox and gives code OS privileges
- Agent swarms and recursion multiply model spend, so budgets and depth limits matter
Install
Manual install
pi install npm:pi-fabricRequires Node.js 24 or newer and Pi 1.0.0 or newer.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agent programs run as the session and call Pi tools, including file writes and commands
- Choosing CPython or the Node or Bun modes removes the sandbox and gives code OS privileges
- Agent swarms and recursion multiply model spend, so budgets and depth limits matter
This is third-party code. Review the repository files before installing.
What it does
pi-fabric is an extension for the Pi coding agent that replaces a set of separate tools with one programmable tool, fabric_exec. The model writes a short program and calls Pi tools, MCP servers, child agents and workflows from it. Branching, loops and parallel calls are written in code, and intermediate results stay in the sandbox and out of the model's context. By default the program is written in TypeScript and runs in an isolated QuickJS after a static type check. A Python (Monty) sandbox can be chosen instead in the settings file. Agents can run as one-shot workers, stay persistent or form a council of several reviewers. Shared tasks and state live in a mesh. Timeouts, recursion depth and a shared budget keep costs in check, and risky steps need approval. Ready-made patterns live in skills: fabric-workflow for parallel audits, fabric-rlm for tasks larger than the context and fabric-council for reviews. The Pi interface gains an activity panel, the `/fabric` command with a dashboard and settings, and a conversation with a child agent.
Who it is for. For Pi users who run large multi-step tasks: audits across dozens of files, migrations and reviews by several agents.
Good fit when
- You need to check every auth file in the repository in parallel and merge verified findings into one report
- The task does not fit the context window and the agent must split it recursively, as in the fabric-rlm skill
- You want a design shown to several independent reviewers and a single decision from the fabric-council skill
- Migration steps must survive restarts and need human approval before release, which fabric-graph handles
- You want one tool instead of ten: fabric_exec calls MCP servers and Pi tools from a single program
Not a fit when
- You work in Claude Code or Cursor: the extension is written for the Pi agent only
- The task is simple and fits a couple of ordinary tool calls: the layer of programs and agents is excess
- The machine has no Node.js 24 and Pi 1.0 and you cannot install them: those are the minimum versions
Example request
Check all files in src/auth in parallel for unsafe token handling and give one report with verified findingsLimitations
Node.js 24 or newer and Pi 1.0.0 or newer are required, and the extension does not work in other agents. The Python (Monty) sandbox is limited: arbitrary imports and third-party packages are unavailable. Full CPython is turned on with a separate `executor.pythonRuntime` setting and runs with OS privileges, needing sandbox-exec on macOS or bwrap on Linux.
A Coral Bricks provider ships with the extension, with sign-in through `/login coralbricks`. The project's maintainers work at that company, and the provider can be excluded with a package filter. For individual MCP servers Pi uses mcporter, and Pi's native MCP is enabled with the `mcp.nativeServers` option.
How to disable. Remove the package from Pi with its package removal command or drop it from the installed list. The ~/.pi/agent/fabric.json settings file can be deleted by hand, and after a Pi restart it returns to ordinary tools.
FAQ
How do I install pi-fabric?
Run `pi install npm:pi-fabric`. You can also install straight from the repository: `pi install git:github.com/fabric-runtime/pi-fabric`. For a local check use `pi -e /absolute/path/to/pi-fabric`.
What does fabric_exec do?
It is the single tool the extension exposes to the model. The model sends a program and Fabric runs it in a sandbox, calling Pi core, MCP servers and agents along the way. Independent calls run in parallel, and only the final object returns to the conversation.
How do I switch to Python instead of TypeScript?
Add `{ "executor": { "kernel": "python" } }` to `~/.pi/agent/fabric.json` or to a trusted project's `.pi/fabric.json`. By default Python runs in the Monty sandbox without file or network access.
Which skills ship with pi-fabric and how do I call them?
For example fabric-guide, fabric-workflow, fabric-rlm, fabric-council, fabric-fusion, fabric-swarm and fabric-graph. Call them explicitly with a command like `/skill:fabric-workflow` and a task description. The model loads them after a direct user request.
How do I watch agents and change settings?
The `/fabric` command opens the activity and topology views, `/fabric settings` shows settings and saves them to fabric.json, and `/fabric chat <agent>` or Ctrl+Shift+A opens a conversation with a child agent.
Discussion
No questions or reviews yet.
Related
PluginsOpen-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
An engineering system for coding agents: 68 agents, 292 skills, hooks, memory, rules and the AgentShield security scanner
DeepSeek Harness is an open-source AI agent from DeepSeek where every feature is a plugin. It runs in the browser, on macOS and on Windows.
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Ollama runs open models on your own machine and serves them to agents through a local API compatible with OpenAI and Anthropic.
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| pi-fabric (this entry) | no | 304 | |
| pi-subagents | no | 1,353 | subagents and Claude Code-style orchestration for Pi; Fabric adds a code mode with one tool, workflows, a mesh and budgets |
| pi-extensions | no | 731 | a set of separate Pi extensions: goals, plan mode, LSP; Fabric is built around one programmable tool |
README in short
The README presents the project as a programmable runtime for tools and agents in Pi and explains how the TypeScript and Python kernels differ. Installation, a skills table with example prompts and documentation links follow.