Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
Learning path
Pick the path that matches what you want to do.
Path A — Run the platform and message an agent (≈ 20 minutes)
- Quickstart —
make minikube-setup, then message the seededchatllmagent from the desktop app or the API - Tour the Desktop App — screens, live activity, in-chat approvals, and artifacts, beyond the one message you just sent. To point it at more than one instance or handle updates, see Desktop setup & updates
- Optional: connect Telegram
- Read Why evenfire
Success looks like: an LLM reply (desktop, API, or Telegram) — and an approval gate firing when you ask the agent to run a real tool.
Path B — Understand the platform (≈ 30–60 minutes)
- Why evenfire and When to use evenfire
- Root README security model
- Architecture overview (skim services + message lifecycle)
- Surfaces index — the three UIs, who each is for, and what each may reach
- CRD reference index
Success looks like: you can explain Host → Context → McpServer and why NetworkPolicies matter.
Path C — Run the full local stack (half day)
- Minikube deploy guide
- Configure approvals
- Add an MCP server
- Track usage & set budgets — meter and cap what agents spend
- WorkflowRecipe CRD reference if you need multi-workload recipes
- Self-host the registry: connect, then publish a plugin
- Give agents files: shared & global files
Success looks like: Control UI up, a Host responding, an approval or connector path exercised.
Path D — Contribute
- Contributing
- Code names
- Per-service README +
npm testin the package you touch - E2E guide only if you change cross-service behavior
Path E — Production planning
- Production deploy notes
- Security model + SECURITY.md
- License (MPL-2.0) in root README and LICENSE
- What's OSS vs the managed service: Open core: self-host vs hosted
Stuck?
See the FAQ.