Mimir Lab builds and teaches AI automation — from n8n workflows to autonomous agents. Open pipelines. Real costs. No fluff.
$ mimir status --tiles
✓ active workflows —
✓ RSS sources —
→ failure rate 7d —
⚠ last error —
→ last brief —
↳ "…"
→ last publish —
updated … · next refresh in 30s
Production-grade n8n workflows + Claude agents that actually run. No toy demos — real pipelines with error handling, retry logic, and Telegram alerts.
Self-hosted pipelines on your own infra. Gmail classifiers, LinkedIn publishers, voice agents — running 24/7 without paying per-task SaaS fees.
Open workflows, real cost breakdowns, and honest write-ups. If we build it, we document it — so you can replicate or improve on it.
From the feed
A direct comparison of n8n 2.0, Zapier, and Make (formerly Integromat) for teams building AI-native automation in 2026. Pricing, AI nodes, self-hosting, and real-world limits.
Featured workflow
Webhook-triggered content brief system. Accepts a raw topic via Telegram, passes it to Claude for research and drafting, checks the draft is grounded in its sources before publishing, generates a structured LinkedIn post, auto-publishes in EN and IT via the LinkedIn API, and notifies back on Telegram. Fully self-hosted on n8n — the real workflow has 20 nodes, viewable in full on /workflows.
The human review gate is a deliberate design choice, not a limitation — documented as a supervised pipeline within the HILC→fully-autonomous spectrum.
Webhook trigger
Telegram message → n8n webhook
Claude drafts
Research + structured post generation
Human review
Draft sent to Telegram for approval
Automated fact-check
A second model verifies the text is grounded in its sources before it can go live
Auto-publish EN/IT
LinkedIn API posts in both languages on confirmation
