Open Source Agent Platform

AI agents that plan, execute, and adapt

A production-ready AI platform with built-in portal, streaming, and DAG visualization. Run standalone or embed into legacy systems via the connector platform — works with any LLM provider.

GitHubDocs
$git clone https://github.com/fim-ai/fim-agent.git
Out of the box
Python 3.11+
Any LLM provider

Everything you need, ready to deploy

From standalone AI assistant to embedded engine for legacy systems.

Dynamic DAG Planning

LLM decomposes goals into dependency graphs, runs steps in parallel, and auto-revises the plan up to 3 rounds.

Out of the Box Portal

Full-featured AI portal with streaming, DAG visualization, dark/light theme, and command palette. Clone, configure, launch.

Connector Platform

Bridge any legacy system — databases, REST APIs, messaging platforms. Auth injection, response extraction, zero code change on their side.

Grounded RAG Pipeline

Hybrid retrieval with RRF, inline [N] citations, conflict detection, and explainable confidence scores.

Pluggable Tool System

Auto-discovered built-in tools: Python executor, web search, shell exec, HTTP request, and MCP protocol support.

Smart Context Engine

LLM-powered compaction, ContextGuard token budgets, and pinned messages — never lose critical context.

Plan. Execute. Verify.

Three-phase agent loop that adapts on the fly.

re-planPlanDAG decomposition1ExecuteParallel tool calling2VerifyAnalyze & re-plan3confidence < 0.5 ?
1

DAG Planner

Decomposes user intent into a dependency graph of executable steps with parallel branches.

2

DAG Executor

Runs steps concurrently, feeding outputs downstream. ReAct agents handle each step with tool access.

3

Plan Analyzer

Evaluates results against the original goal. If confidence is low, triggers automatic re-planning.

Quick Start

git clone https://github.com/fim-ai/fim-agent.git
cd fim-agent
cp example.env .env  # add your API key
./start.sh           # launch portal

How FIM Agent compares

Built different. Built lighter.

FIM AgentDifyManusCoze
ApproachAgent platform + sidecarLow-code platformAutonomous browser agentVisual bot builder
PlanningDynamic DAG + auto re-planningStatic workflowTask decompositionFixed flow
Legacy integrationConnector platform (DB + API + push)Limited API nodesBrowser automationPlugin marketplace
Built-in portalStreaming + DAG viz + themesFull UICloud onlyCloud UI
Self-hostedYes, single processYes, Docker stackCloud onlyYes (Docker)
Human confirmationWrite ops require approvalNoNoNo