For Developers

Automation that behaves like software.

Flows you can pull to disk, diff in git, test against mocks, and ship. A REST API for everything. An MCP server so AI agents can build alongside you. All on infrastructure you control.

Flows as Code

Pull. Edit. Push.

Your flows, on disk, under version control.

Folder sync mirrors your flows to disk as clean, diffable JSON. Pull them into a repo, review changes like any other code, and push them back — with a per-flow report of exactly what changed on either side.

Your automation gets branches, reviews, and history. No more "who changed this flow, and why?"

  • Diffable JSON definitions for every flow
  • Per-flow push & pull status reports
  • Conflict and missing-reference detection
  • Works with any git workflow you already run
Folder sync
GET /flow/folder/status
~ invoice-processing modified (remote)
+ vendor-onboarding new (local)
POST /flow/folder/pull
✓ 42 flows synced → ./flows
git diff flows/invoice-processing.json
POST /flow/folder/push
✓ 2 flows updated · 0 conflicts

Break it in the sandbox. Not in production.

Flows are software — so they get a test story, not a "run it and pray" button.

🧪

Mocked sandbox

Run flows against mocked integrations in an isolated container environment. No live credentials, no side effects, no accidental emails to real customers.

Test cases in the editor

Define test cases right next to the flow, run them on demand, and catch regressions before a change reaches a real execution.

⏮️

Replay & inspect

Re-run past executions, switch between runs, and inspect every step's inputs and outputs — debugging with real data instead of guesses.

MCP

AI agents are first-class engineers here.

flow8 ships an MCP server. Point Claude, ChatGPT, or Cursor at it and agents can scaffold new flows, modify existing ones, and validate drafts — using the same folder sync and sandbox you use.

It works the other way too: every flow you ship becomes a tool an AI assistant can call. Your business logic, available to every agent your team runs — with access rules deciding who may call what.

Claude ChatGPT Cursor Claude Code ChatGPT Codex
Your AI agent
MCP Server
Flow definitions
Sandbox
Live runs
Agents author, test, and invoke flows — audited like any user

Everything has an API.

No SDK lock-in, no black boxes — the surfaces your team needs to build on, extend, and observe.

🔌

REST for everything

Flows, runs, logs, and admin — every operation is a standard HTTP call from any language.

🔑

API keys

Scoped keys for service-to-service auth, created and rotated in the UI.

📡

WebSocket streams

Live execution state pushed to your clients the moment steps complete. No polling.

🧩

Custom Go modules

One Go file, auto-discovered — your proprietary logic appears in the visual builder immediately.

🗄️

Your databases

Embedded F8DB out of the box, MongoDB and PostgreSQL natively — plus Qdrant and Weaviate for vector workloads.

📦

Container deploy

Multi-arch images (amd64/arm64) for on-premise, private cloud, or air-gapped environments.

🔐

Passkeys & access rules

WebAuthn login, granular per-resource permissions, and user groups — enforced at the API layer.

📜

Audit middleware

Every API call logged with actor, IP, and sanitized payloads. The trail can't be switched off.

Pull a flow. Diff it. Push it back.

Book 30 minutes with an engineer. We'll stand up flow8 in a container, sync a flow to disk, and let your AI tooling loose on it.

REST API · MCP server · Multi-arch containers