Skip to content
A real fuseraft REPL session: asking the agent what the !<command> shell escape does, then using it to run git log directly Open source

fuseraft

Chat in your terminal. Scale into multi-agent pipelines when one agent isn't enough.

Get Started GitHub

What it does

Run fuseraft and start chatting with a model that can read and edit files, run shell commands, search your codebase, and use git — no config file needed. When one agent isn't enough, fuseraft scales into declarative, multi-agent pipelines defined in YAML, where routing validators mechanically enforce that each agent did what it claimed before the pipeline advances. Built on Microsoft Agent Framework.

  • Terminal AI assistant

    fuseraft repl is an interactive chat session with a single model — safe by default (HITL approval, filesystem/shell/git sandbox), resumable, with ~50 slash commands and a !<command> shell escape for running things yourself without leaving the chat.

    REPL

  • Agent teams as YAML

    Define each agent's name, model, system prompt, and plugins in a single YAML config. The coordinator routes work between them automatically.

    Configuration

  • Model-agnostic

    Mix frontier LLMs and local SLMs per agent in the same team — Anthropic, OpenAI, Google, Mistral, xAI, DeepSeek, Azure OpenAI, or any model served through Ollama. Rotate API keys automatically on rate limits.

    Models & Providers

  • Rich plugin ecosystem

    Every agent can call filesystem, shell, git, HTTP, JSON, search, and Docker sandbox tools out of the box. Connect any external MCP server.

    Plugins

  • Resilient sessions

    Sessions checkpoint after every turn. Interrupt anytime and resume exactly where you left off — no work is lost.

    Sessions

  • Spec-driven development

    Use --spec to anchor the team to an agreed specification before implementation begins. Routing validators block handoffs until evidence is present.

    Spec-Driven Development

  • Governance & cost control

    Track token usage and estimated cost per turn. Enforce hard spending caps. Apply execution rings, prompt injection detection, and a hash-chain audit log.

    Governance

Quick start

curl -fsSL https://raw.githubusercontent.com/fuseraft/fuseraft-cli/main/install.sh | bash

Then run the setup wizard on first launch:

fuseraft

No configuration found at ~/.fuseraft/config

Provider setup
Configure your provider and API key, then pick a model.

Provider URL  (http://localhost:11434): https://api.anthropic.com
API Key       (leave blank for Ollama): ••••••••

Model  (2 available from https://api.anthropic.com)
> claude-sonnet-4-6
  claude-opus-4-6

>

That's it — you're chatting, with file, shell, search, and git tools already available.

irm https://raw.githubusercontent.com/fuseraft/fuseraft-cli/main/install.ps1 | iex

Then run the setup wizard on first launch:

fuseraft

When one agent isn't enough, generate a team config and run a multi-agent pipeline instead:

fuseraft init
fuseraft run -c .fuseraft/config/orchestration.yaml "Add a hello-world endpoint to this project"

Full installation guide · REPL walkthrough

Documentation

Doc What it covers
Getting Started Prerequisites, installation, first run
REPL Interactive terminal chat — tools, safety model, ! shell escape, skills, sessions
Writing Effective Tasks Task descriptions that produce correct, verifiable results
Spec-Driven Development Using --spec to anchor agents before implementation begins
CLI Reference All commands and flags
Scripting & Automation Running fuseraft from bash/Python, --json output, event-driven pipelines
Configuration Full config schema (YAML and JSON)
Models & Providers Model configuration and auto-detection
Plugins All built-in tools agents can call
Strategies Selection and termination strategies
Routing Validators Anti-hallucination handoff guards
Harness Engineering Building configs that enforce correctness mechanically
MCP Integration Connecting external MCP servers
Security & Sandbox File and network containment
Governance Execution rings, audit log, circuit breaker, SLO tracking
Evals Running agent teams against scored test cases; CI integration
Sessions Resumption, HITL, cost tracking, compaction
Context Management How fuseraft manages context across a long session
Context Store Importing reference material for agents
Knowledge Layer ADRs, graph, provenance
Skills Portable skill packages and cross-session skill index
Examples Ready-to-use config examples
Design Architecture, layer map, MAF usage, and decision log

VS Code Extension

The Fuseraft VS Code extension brings the CLI into your editor — run tasks, browse sessions, validate configs, and get YAML/JSON IntelliSense, all from the activity bar.