bytechefhq/bytechef

Open-source AI agents and workflow automation. Self-host or embed in your SaaS. Apache 2.0 alternative to n8n and Zapier.

ByteChef – Open‑source AI‑agent + workflow automation platform

What it is ByteChef is a self‑hostable platform that combines two traditionally separate worlds:

AI agents – a drag‑and‑drop component that runs the full “agent loop” (choose a model, select tools, execute, observe, decide next step) with streaming or structured output.
Workflow automation / iPaaS – a visual editor for building, version‑controlling, and running complex integration flows.

The platform lets you embed an agent as a single step inside a workflow, and conversely expose any workflow as a tool that agents (or external MCP servers) can call. All of this runs on a durable, PostgreSQL‑backed runtime that can be scaled horizontally.

Core capabilities (as described in the README)

Category	What you get out‑of‑the‑box
AI agents	Built‑in agent component with full loop, streaming output, and a UI editor.
LLM providers	Direct connectors for 12 providers (OpenAI, Anthropic, Azure OpenAI, Bedrock, Gemini, Mistral, Grok, DeepSeek, Nvidia, Perplexity, Stability, Ollama) plus aggregator components (OpenRouter, LiteLLM, NanoGPT).
Tools / Connectors	>250 pre‑built connectors (CRM, marketing, e‑commerce, cloud storage, databases, help‑desk, finance, etc.). Each connector is also an agent‑tool and can be exposed via MCP.
Memory back‑ends	8 options: built‑in, JDBC, Redis, MongoDB, Cassandra, Neo4j, vector‑store‑backed, in‑memory.
RAG / Knowledge bases	Native ingestion & chunking, 15+ vector stores (pgvector, Pinecone, Qdrant, Weaviate, Milvus, Couchbase, Neo4j, Redis, Typesense, MariaDB, Oracle, S3, built‑in). Two ready‑made RAG patterns (`rag-modular`, `rag-questionanswer`).
Guardrails	12 ready‑to‑attach safety checks (PII, jailbreak, NSFW, topical alignment, keyword filters, secret‑key redaction, URL sanitisation, custom regex/rules, etc.).
Workflow features	Visual canvas with JSON view, condition/branch/loop/map/parallel/fork‑join, triggers (webhooks, polling, schedule, app events), polyglot code steps (JS, Python, Ruby via GraalVM), durable execution, queue‑mode scaling (Redis, RabbitMQ, Kafka, SQS, etc.), Git‑native versioning, “workflow‑as‑API” endpoints.
Copilot	Natural‑language UI that generates a workflow (including an AI‑agent step) from a plain‑English prompt, explains failures and suggests fixes (Enterprise edition).
Enterprise‑only extensions	Workflows‑as‑APIs, Git‑native promotion, micro‑service deployment, SSO/SAML/OIDC, SCIM, advanced RBAC, AI Gateway (model routing, quotas, cost controls), embedded iPaaS for SaaS products.

Typical use cases

Scenario	How ByteChef helps
Customer‑service automation	An AI‑agent step decides whether to answer directly, look up a knowledge‑base via RAG, or route to a human via Slack/email approval.
Order‑processing with approvals	A workflow orchestrates payment, inventory checks, and shipping; an agent can decide on discounts or refunds, while a human‑in‑the‑loop step pauses for manager approval.
Data‑enrichment pipelines	Use connectors to pull data from a CRM, call an LLM to generate missing fields, store results back, all within a single visual flow.
Embedding AI into a SaaS product	Deploy ByteChef as a microservice, expose specific workflows as MCP tools, and let your product call them as “AI features” without building the orchestration layer yourself.
Rapid prototyping	Write a one‑sentence prompt to the Copilot, get a full workflow with an agent and necessary connectors generated instantly.

Getting started (quick‑start)

Prerequisite – Docker Desktop (or any Docker engine).

Download the compose file

curl -O https://raw.githubusercontent.com/bytechefhq/bytechef/master/docker-compose.yml

Launch
```
docker compose -f docker-compose.yml up
```
This starts PostgreSQL and the ByteChef server.
Open a browser at http://localhost:8080/login, create an account, and you’re in the visual editor.
Follow the “Build your first agent in 60 seconds” checklist in the README to add a trigger, drop an AI Agent component, pick a model, attach a connector as a tool, and test the workflow.

For environments where Docker‑Compose isn’t available, the README also provides step‑by‑step commands to create a dedicated Docker network, run a PostgreSQL container, and start the ByteChef container manually (with a persistent ~/.bytechef volume for the encryption key).

License & community

Core code – Apache 2.0 (free for commercial use, modification, and redistribution).
Enterprise extensions – ByteChef Enterprise License (located under /ee/).
Support – Docs site, Discord community, GitHub issues, and a public roadmap board.
Contributing – A CONTRIBUTING guide is provided; the project welcomes pull requests for new connectors, bug fixes, and feature enhancements.

Bottom line

ByteChef is an open‑core platform that lets teams build, run, and manage AI‑driven automation without stitching together separate LLM libraries, workflow engines, and connector code. By treating an AI agent as a first‑class step in a visual workflow, it provides deterministic control (branching, retries, approvals) alongside the flexibility of large‑language‑model reasoning, all under a single audit log.