Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
When to use evenfire
evenfire is a self-hostable platform for multi-channel LLM agents that take real actions — with approvals, permissions, and audit-oriented controls built in. This page describes categories of tools so you can choose the right fit. It does not rank or name competing products.
At a glance
| evenfire | Personal assistants | In-process agent frameworks | |
|---|---|---|---|
| Primary job | Governed agent platform | Assistant you message day-to-day | Library to build agent features |
| Install metaphor | Kubernetes-native (make minikube-setup locally) |
Single process / gateway | pip / npm into your app |
| Config model | Kubernetes CRDs + control plane | Config files / CLI / skills | Code-first graphs and crews |
| Multi-channel | Telegram, email, Slack + desktop | Many chat surfaces (often the focus) | Usually your application’s UI |
| Tool governance | Context allowlists + NetworkPolicies + approvals | Approvals / allowlists / sandbox (varies) | Application-defined |
| Fleet / org policy | Hosts, contexts, recipes as reviewable config | Usually single-user or small team | Your responsibility |
| Best first hour | Own an agent runtime; then grow into policy | Chat tonight with minimal infra | Embed an agent loop in code |
Choose a personal assistant when…
- You want chat surfaces (Telegram, Discord, etc.) with minimal infrastructure
- Skills, memory, and “message it from your phone” matter most
- A single-user or single-node security model is enough
Choose an in-process framework when…
- Agents are a feature inside your product
- You need custom multi-step graphs or research experiments in code
- You will own deployment, identity, and tool isolation yourself
Choose evenfire when…
- Agents must call real tools against real systems and you need policy
- You want default-deny networking and human approval as platform primitives, not bolt-ons
- Multiple hosts, channels, connectors, and workflows should be declarative and reviewable
- You will self-host on infrastructure you control (Kubernetes — minikube locally, any cluster in production)
Overlap (honest)
- evenfire also does multi-channel messaging and MCP tools — so do personal assistants.
- evenfire includes an agent runtime (
mcp-host) — it is not “only CRDs.” - Frameworks can be composed into production systems; evenfire is closer to the control plane those systems need once actions are real.