Open Source · AGPL-3.0 · Production Ready

AI OrchestrationThat ActuallyShips.

Wire agents, tools, and 14+ LLMs into deterministic pipelines

bash
$ curl -sSL https://raw.githubusercontent.com/synapseorch-ai/synapse-ai/main/setup.sh | bash
0+
LLM Providers
0
Step Types
0
Messaging Platforms
Jira to Deploy Pipeline

Jira to Deploy Pipeline

End-to-end pipeline that takes a Jira ticket from planning to a merged, deployed PR.

14+ Model Providers · 5 Messaging Apps · One Platform

Claude (Anthropic)
Claude (Anthropic)
ChatGPT / OpenAI
ChatGPT / OpenAI
Google Gemini
Google Gemini
Grok (xAI)
Grok (xAI)
DeepSeek
DeepSeek
Ollama (Local)
Ollama (Local)
AWS Bedrock
AWS Bedrock
Slack
Slack
Discord
Discord
Telegram
Telegram
Teams
Teams
WhatsApp
WhatsApp
Claude (Anthropic)
Claude (Anthropic)
ChatGPT / OpenAI
ChatGPT / OpenAI
Google Gemini
Google Gemini
Grok (xAI)
Grok (xAI)
DeepSeek
DeepSeek
Ollama (Local)
Ollama (Local)
AWS Bedrock
AWS Bedrock
Slack
Slack
Discord
Discord
Telegram
Telegram
Teams
Teams
WhatsApp
WhatsApp
Why Synapse AIEnterprise Ready · Self-Hostable

Built different, on purpose

Most AI tools are prototypes dressed up as platforms. Synapse is built for businesses that need to ship reliable AI solutions — not experiment forever.

Deterministic by Design

Workflows run as strict DAGs. Every execution follows a predictable path. No hallucinated tool chains, no surprise loops.

Right Model, Every Step

Mix cheap models for routing with powerful ones for reasoning. Optimize cost and performance at each node, not across the whole workflow.

If You Can Write It, Agents Can Use It

Python functions, REST APIs, webhooks, MCP servers: any tool you already have works out of the box. Zero glue code.

Runs Where You Need It

Deploy on-premises with Docker, on any cloud provider, or locally with Ollama. Your data stays in your infrastructure. No vendor lock-in, no surprise egress.

Platform Features

Everything you need to ship AI workflows

Not a framework with toy examples. A complete platform for production AI workflows.

Visual DAG Builder

Drag-and-drop orchestration canvas powered by ReactFlow. 10 step types: Agent, LLM, Tool, Evaluator, Parallel, Merge, Loop, Human, Transform, and End.

10 step types

14+ LLM Providers

Mix models per step: cheap models for routing, powerful models for reasoning. Claude, GPT-4, Gemini, Grok, DeepSeek, Ollama local, AWS Bedrock, and more.

Per-step override

Real Tool Servers

14+ built-in tool servers: Docker Python sandbox, Vault, SQL Agent (Postgres/MySQL/SQLite), Browser (Playwright), stealth Web Scraper, PDF/Excel parsers, and MCP.

Docker sandboxed

AI Builder

Describe your orchestration in plain English. The AI builds the full DAG for you: agents, connections, step configs. Then you run it. No canvas required.

Chat-to-DAG

Human-in-the-Loop

Human steps pause execution mid-workflow and wait for approval. Resume from the UI, Slack, Telegram, or any connected channel. Runs survive server restarts.

Resumable execution

Messaging Integrations

Deploy to Slack, Discord, Telegram, Teams, or WhatsApp. Per-channel agent binding, multi-agent chat mode, scheduling, and result auto-push.

5 platforms

Architecture

How Orchestration Works

Every orchestration is a strict DAG. Execution follows your exact design. No surprises, no hallucinated detours.

01

Connect MCP Servers

MCP

Plug in any MCP-compatible server to extend your orchestration with external tools and services. Browse the built-in registry or paste a server URL, and new capabilities appear instantly in every agent.

02

Add Custom Tools

Tool

Paste your Python handler directly into the editor or drop in an API webhook URL. No SDK needed. Define the schema, inputs, and secrets in one form and your tool appears instantly in every agent's toolbox.

03

Define Your Agents

Agent

Create specialized agents in the settings panel with custom system prompts, tool access, and per-agent model overrides. Each agent has a focused role.

04

Wire the DAG

Parallel

Drag nodes onto the canvas and connect them. Choose from 10 step types: Agent, LLM, Tool, Evaluator, Parallel, Merge, Loop, Human, Transform, or End.

05

Set Per-Step Models

LLM

Override the LLM at every step. Use a cheap model for routing decisions, a powerful model for complex analysis. Optimize cost without sacrificing quality.

06

Run the Orchestration

Tool

Execution follows your exact DAG. No hallucinated detours. Each step processes input from shared state, produces output, and passes it to the next node.

07

Human Review & Resume

Human

Human steps pause execution and notify you via Slack, Telegram, or the UI. Approve, reject, or provide feedback. Execution continues exactly where it stopped.

Execution Flow
Connect MCP Servers
running
Add Custom Tools
pending
Define Your Agents
pending
Wire the DAG
pending
Set Per-Step Models
pending
Run the Orchestration
pending
Human Review & Resume
pending
Progress14%

All step types

AgentLLMToolEvaluatorParallelMergeLoopHumanTransformEnd

Why Synapse

Not a framework. A platform.

Most tools solve one piece. Synapse AI ships the whole thing.

FeatureSynapse AILangChainCrewAISuperAGIAutoGenn8n / ZapierLangflowFlowiseDify

Visual DAG Canvas

Drag-and-drop nodes with real-time per-step status, no YAML files or code-first setup

Local LLM Support

14+ providers including Ollama; run fully air-gapped or mix local and cloud models in the same workflow

Human-in-the-Loop Gates

Pause mid-run for human approval; resume from browser or connected Slack/Telegram/Teams. Survives server restarts.

Docker Code Sandbox

Executes AI-generated Python in isolated containers with data-science libraries and a persistent cross-session file vault

Stealth Web Scraping

Headless Chromium with anti-bot evasion. Scrapes LinkedIn, JS-heavy, and financial sites straight out of the box.

AI-Generated DAGs

Describe a workflow in plain English; an AI builder drafts the full DAG, creates sub-agents, and asks for approval before saving

Per-Step Model Override

Assign a different LLM to each node: fast local model for routing, Claude Opus only where deep reasoning actually matters.

Nested Orchestrations

Promote any workflow to a callable agent and stack pipelines on top of each other like composable, reusable functions

Full Observability

Per-step token counts, live USD cost enforcement, p95/p99 latency profiles, and complete execution history per run

Messaging Integration

Bind Telegram, Slack, Discord, Teams, or WhatsApp to any workflow; human gates push approval prompts directly to your chat

Pipelines as Callable Agents

Promote any orchestration into a reusable agent tool — call a full pipeline from inside another pipeline like a composable function

Typed Shared State

All steps read/write a single typed schema dict — no edge wiring between nodes. State persists across agents, transforms, and human pauses in one run.

Deterministic Control-Flow Steps

IF/Else, Switch, Transform (Python), Extract JSON — native zero-LLM step types that branch, reshape, and route data without spending tokens

Yes Partial No

Use Cases

What people build with Synapse

Real workflows, real production use cases. Not toys.

Content Research & Writing

Research any topic online, synthesize findings, draft long-form content, create a Google Doc, and share the link. End to end, no manual steps.

AgentToolEvaluator

Code Review & Autonomous PR

Review pull requests, write fixes, run tests in a sandbox, and merge when ready. Human gate at each stage ensures you stay in control.

AgentHumanTool

Market Intelligence Report

Scrape financial sites with stealth mode, aggregate and analyze data with a Python sandbox, then auto-push a summary to Slack on a schedule.

ToolTransformLLM

Customer Support Bot

Multi-agent Slack or Telegram bot that classifies, routes, and answers. Users switch agents mid-conversation. Escalates to human when confidence is low.

AgentEvaluatorHuman

Database Q&A Agent

Natural-language SQL: describe what you want, get structured results back. Works with PostgreSQL, MySQL, and SQLite. No SQL knowledge required.

ToolLLMTransform

Deep Web Research Agent

Navigate websites, extract data from PDFs and JS-heavy pages with stealth mode, store findings in Vault, and produce a fully cited research report.

AgentToolLoop

If you can wrap it in an API, Python function, or MCP server? Synapse can automate it.

Agents and orchestrations are not limited to what's built in. Any tool you already have becomes a first-class step in your workflow.

See It In Action

Watch Synapse AI work

Full UI walkthrough, from creating agents to running a multi-step orchestration.

Synapse AI: Platform Demo

Multi-agent orchestration in action

AI Builder in Action

Describe your workflow in chat. Watch the full DAG appear in seconds.

Human Review Gate

Execution pauses at the Human step. You review, approve, and it continues.

Real-Time Step Status

Watch each node light up as it completes. Live cost and turn tracking.

Examples

Don't start from scratch

Start with a working example and make it yours. Every example is a fully configured bundle of agents, orchestrations, and MCP servers, ready to customize.

Starter Example

Get up and running fast. Includes a Personal Assistant with full tool access and a Web Research Agent, plus a ready-to-use Content Research & Writing Pipeline.

5 agents
2 orchestrations
  • Personal Assistant with full tool access
  • Web Research Agent
  • Content Research & Writing Pipeline
  • Browser + Vault + Sandbox tools included

Developer Example

Built for developers. Includes a Code Review Agent, Software Engineer Agent, QA Engineer, Deployment Monitor, and a complete Dev and Strict Pipeline orchestration.

7 agents
3 orchestrations
2 MCP servers
  • Code Review & Software Engineer Agents
  • QA Engineer + Deployment Monitor
  • Dev Pipeline + Strict Pipeline orchestrations
  • GitHub + Filesystem MCP servers

Productivity Example

For business and content power users. Includes a Data Analyst, Content Writer, Jira Analyst, Slack Notifier, and Market Intelligence orchestration.

4 agents
1 orchestrations
3 MCP servers
  • Data Analyst + Content Writer Agents
  • Jira Analyst + Slack Notifier
  • Market Intelligence orchestration
  • Jira + Slack + Notion MCP servers

OneCommandtoInstall.

macOS, Linux, Windows, or Docker. Deploy on your infrastructure — on-premises, cloud, or local.

bash
$ curl -sSL https://raw.githubusercontent.com/synapseorch-ai/synapse-ai/main/setup.sh | bash
Then use the CLI
synapse startStart backend + frontend, open browser
synapse stopStop background processes
synapse upgradePull latest and rebuild
synapse uninstallRemove everything

CLI requires Python 3.11+ and Node.js 22+. Docker image is self-contained — no prerequisites needed.