Automations that write, deploy, and repair themselves.
Forge is a self-hosted AI operations platform. Describe what you want in plain English — an agent layer generates a valid workflow, validates it, deploys it to n8n, watches every run, and fixes failures before you notice.
No login. The demo runs on realistic sample data.
you › watch for new ML job posts daily and draft cold emails to the hiring managers
- ✓ generated workflow ml-jobs-outreach · 5 nodes
- ✓ validated node types, connections, credentials
- ✓ deployed to n8n · version v1 · scheduled daily 09:00
- ◷ monitoring runs — auto-repair armed
Built on a production stack
Six systems that turn intent into running, monitored automation.
n8n executes workflows. Forge is the brain on top — it decides what to build, whether it's safe, and what to do when it breaks.
Natural-language generator
Describe an automation in a sentence. The agent turns it into schema-valid n8n workflow JSON — never hand-edited in the UI.
Static validator
Every generated workflow is checked before deploy: node types exist, connections resolve, credentials are referenced right, and destructive actions are blocked without an explicit approval flag.
Versioned registry
Each deploy is an immutable version. Rollback to any previous one is a single API call — with a backup and audit entry written first.
Run monitor
Executions are ingested and health is computed per workflow — success rate, streaks, last-run status — so failing automations surface instantly.
Self-healing diagnostician
On failure it pulls logs plus the workflow definition, writes a root-cause analysis, and proposes a patch. Low-risk fixes auto-apply; risky ones wait for you.
LLM gateway
Every model call flows through one module: cost + token logging, a hard daily budget, exponential-backoff retries, and a local Ollama fallback.
One instruction in. A self-operating workflow out.
The agent layer runs a closed loop. Generation happens as a queued job, so the API never blocks — and the loop keeps turning after you've walked away.
Describe
You type an instruction in plain English from the dashboard chat box.
Generate
A background job turns it into n8n workflow JSON via the LLM gateway.
Validate
Static checks confirm the graph is valid and nothing destructive slips through.
Deploy
It's pushed to n8n as an immutable version, ready to run on schedule or webhook.
Monitor
Every run is ingested; health, success rate, and failure streaks are tracked.
Repair
On failure the diagnostician finds the cause and patches it — or asks you first.
Four clean layers. The engine is swappable.
Built stateless and horizontally scalable from day one: the API holds no state, generation runs off a Redis queue, and no n8n-specific logic leaks outside the adapter.
Dashboard
Next.js 14 · TypeScript · TailwindThe control surface — workflow health, run timelines, the incident feed, cost charts, and the chat box that drives the agent.
Agent layer
FastAPI · Pydantic v2 · Python 3.11Stateless brain: generator, validator, run-monitor, diagnostician. Scales horizontally because all state lives below it.
Data layer
PostgreSQL 16 · Redis 7Versioned workflow registry, run history, incidents, the full LLM cost log, and an audit trail — plus Redis for the job queue and cache.
Execution engine
n8n (self-hosted)The runtime, driven only through its REST API and isolated behind an engine adapter — so it's swappable without touching the brain.
The whole thing runs in your browser — no backend required.
The live demo is seeded with realistic workflows, runs, incidents, and cost data so you can click through every surface exactly as it behaves in production.
Workflows
12
Open incidents
2
Spend today
$3.41 / $8.00
ml-jobs-outreach
v3 · active
rss-digest-email
v1 · active
hourly-health-check
v5 · active
webhook-slack-alert
v2 · active
LLM spend · 14 days
Forge is one project. There are 18 more — and a few jet engines — over at charanreddy.dev.
Have an idea? Let's talk.
Want to build something — or break something interesting? I'm around. The fastest path is booking a call; the form works too.