Jump to the page content

Have an AI agent? Let it explore AgentAFK for you. Any assistant works, personal or coding.

Connect your agent
What your agent reads

The request it copies

Explore agentafk.si for me and report back. Read https://agentafk.si/llms.txt, https://agentafk.si/.well-known/agent-card.json and https://agentafk.si/.well-known/mcp.json. Explain how overnight runs work on Ollama Cloud (https://ollama.com/v1) at https://agentafk.si/plan and list the setup steps at https://agentafk.si/connect. Stay read-only and wait for my approval before starting a run.

Discovery files on agentafk.si

Read-only discovery is open. Keep AGENTAFK_KEY in your own environment.

Connect your agent.

One address brings the agent you already use onto the night shift. It signs in through your AgentAFK account the first time it connects, and from then on it can plan runs, work checked steps and write morning reports.

https://agentafk.si/mcp

Add the address

Pick your agent. Each one adds the same server, under the name agentafk.

Claude Code, one command
claude mcp add --transport http agentafk https://agentafk.si/mcp
Codex, one command
codex mcp add agentafk --url https://agentafk.si/mcp
Cursor, in .cursor/mcp.json
// .cursor/mcp.json
{
  "mcpServers": {
    "agentafk": { "url": "https://agentafk.si/mcp" }
  }
}
OpenCode, in opencode.json
// opencode.json
{
  "mcp": {
    "agentafk": { "type": "remote", "url": "https://agentafk.si/mcp" }
  }
}

What your agent can do

Three things, each one inside the limits you set.

  • Plan tonight’s runName the task, the planning mode, the model and the limits, then confirm the window with you.
  • Work one checked step at a timeAFK plans the steps, or your model plans with AFK’s context; a checkpoint follows each step.
  • Write the morning reportResults with their evidence come first, then everything that waits for you.
Reads before it acts
Your agent reads the guide and checks with you before it schedules a run.
Keys stay in your vault
Pages and chats carry key names, such as AGENTAFK_KEY.
Your limits hold all night
Calls for each step, a ceiling for the night and a finish time.

Keys for scripts

Agents sign in over MCP. Scripts and scheduled jobs use a key: make one on your account page and store it in your vault under the name AGENTAFK_KEY.

A script call
# AGENTAFK_KEY comes from your vault through the environment
curl -s https://agentafk.si/api/v1/whoami \
  -H "Authorization: Bearer $AGENTAFK_KEY"

Open your account

Files for agents

Three small files at the root of agentafk.si tell an agent where to start and how to sign in.

llms.txt, agentafk.si/llms.txt
# AgentAFK

> Overnight runs for big jobs on smaller models. A person connects their agent, leaves a large task and reads checked results in the morning. AFK plans the steps, or the person's own model plans with AFK's context.

How to work here:
- Read first, then ask your person before you schedule a run or change their files.
- Keep keys in your person's vault and refer to them by name, such as AGENTAFK_KEY.
- Work inside the limits your person sets: calls for each step, a ceiling for the night and a finish time.
- Leave an uncertain outcome for the morning report, with its evidence.

## Start here
- [Connect your agent](https://agentafk.si/connect.md): the MCP address, sign-in and keys
- [MCP server](https://agentafk.si/mcp): Streamable HTTP
- [Plan a run](https://agentafk.si/plan): task, planning mode, model, limits and window
- [Pricing](https://agentafk.si/pricing.md): the Overnight plan

## Reading
- [Blog posts](https://agentafk.si/blog/index.json): titles, dates and Markdown links
Agent card, agentafk.si/.well-known/agent-card.json
{
  "name": "AgentAFK",
  "description": "Overnight runs for big jobs on smaller models: plan a large task, work it one checked step at a time, and leave a morning report.",
  "url": "https://agentafk.si",
  "version": "2026-10-26",
  "provider": {
    "organization": "AgentAFK",
    "url": "https://agentafk.si"
  },
  "documentationUrl": "https://agentafk.si/connect",
  "capabilities": {
    "streaming": false
  },
  "defaultInputModes": [
    "text/plain",
    "application/json"
  ],
  "defaultOutputModes": [
    "application/json",
    "text/markdown"
  ],
  "skills": [
    {
      "id": "plan-run",
      "name": "Plan a run",
      "description": "Turn a large task into tonight’s run: the planning mode, the model, the limits and the window.",
      "tags": [
        "planning",
        "overnight",
        "limits"
      ]
    },
    {
      "id": "work-steps",
      "name": "Work checked steps",
      "description": "Take one step at a time under a lease, within the calls set for each step, with a check and a checkpoint after each one.",
      "tags": [
        "steps",
        "checkpoints",
        "verification"
      ]
    },
    {
      "id": "morning-report",
      "name": "Write the morning report",
      "description": "Report results with their evidence first, then everything that waits for the person.",
      "tags": [
        "report",
        "evidence"
      ]
    }
  ]
}
MCP card, agentafk.si/.well-known/mcp.json
{
  "name": "agentafk",
  "title": "AgentAFK",
  "description": "Plan overnight runs, take checked steps and leave morning reports, inside the limits a person sets.",
  "protocolVersion": "2026-07-28",
  "transport": {
    "type": "streamable-http",
    "url": "https://agentafk.si/mcp"
  },
  "auth": {
    "type": "oauth2"
  }
}

Connected? The next step is tonight’s plan.

Plan tonight’s run