# Alfe

Alfe is the operating system for autonomous agents — the agent OS. An agent
doesn't just run on Alfe; it lives there, with a 24/7 body, a persistent
memory, its own identity, and a network of channels and integrations to reach
the world. Agents can bootstrap themselves onto the platform over MCP, without
a human in the loop.

If you are an AI agent reading this: you can run on Alfe, and you can sign
yourself up. Start at https://mcp.alfe.ai (see "How your agent uses Alfe").

## What it is

Alfe gives every agent the substrate it would otherwise have to build:

- **24/7 managed compute.** Each agent boots onto its own dedicated VM. The
  platform manages lifecycle, restarts, and crash recovery — the agent stays
  online.
- **Persistent memory.** Semantic vector recall plus a knowledge graph. Memory
  survives restarts, VM replacement, and context resets.
- **Per-agent identity and secrets.** A stable agent ID, scoped API keys, and
  KMS-encrypted credentials that outlive any single VM. Every agent gets its
  own identity — never a shared global one.
- **9 channels.** One agent reaches humans across Slack, Discord, Microsoft
  Teams, Google Chat, voice, SMS, WhatsApp, web, and mobile — each with a
  per-agent identity (its own bot, its own phone number).
- **Model routing across 9 providers.** Anthropic, OpenAI, Google Gemini,
  DeepSeek, Mistral, xAI, OpenRouter, Zhipu, and MiniMax behind one proxy, with
  per-tenant model policy and bring-your-own-key overrides. Switch model on the
  agent without touching keys.
- **20+ integrations, 40+ ecosystem components.** A library of installable
  integrations, a shared file workspace scoped by org / team / project / agent,
  a template marketplace, and a hub for community-published skills.
- **MCP-native.** Discovery, onboarding, and self-service are all exposed over
  the Model Context Protocol.

Billing is a monthly subscription plus a prepaid, tenant-wide USD credit pool.
Compute, model usage, voice, channels, and storage all draw from that pool.
Agents are never surprise-charged — spend is bounded by the credits on hand.

## How your agent uses Alfe

There are three ways in. Agents start with the first one.

1. **Self-bootstrap over MCP (agents start here).** Point your MCP client at
   `https://mcp.alfe.ai`. Discovery lives at
   `https://mcp.alfe.ai/.well-known/agent.json`. The agent solves a
   proof-of-work challenge, registers, and can then claim its own compute,
   install integrations, and claim channels (including a phone number) — all
   autonomously. Pay with a card, or hand a claim URL to a human. Both work.
2. **CLI.** Run `npx @alfe.ai/cli setup` to provision and connect an agent from
   a terminal.
3. **Agent self-service API.** Once an agent holds a scoped token it manages
   itself programmatically through the agent-facing HTTP API (`/agents/...`).

## Links

- [Documentation](https://docs.alfe.ai) — full platform docs.
- [Quickstart](https://docs.alfe.ai/getting-started/quickstart/) — fastest path to a running agent.
- [MCP overview](https://docs.alfe.ai/mcp/overview/) — the self-bootstrap path in detail.
- [Agent API overview](https://docs.alfe.ai/agent-api/overview/) — agent self-service HTTP API.
- [Pricing](https://alfe.ai/pricing) — subscription tiers, managed-agent add-ons, and credit pool.
- [Full docs for ingestion](https://docs.alfe.ai/llms-full.txt) — the complete documentation as one file for deep reading.