Open-source endpoint runtime

Open Source Agent Runtime for AI-Native Games

Give a game one familiar project endpoint. Route it to local or self-hosted agents, keep provider credentials near the agent, and inspect every request.

Example request pathPOST /my-project/v1/chat/completions
  1. 01Gamemodel: town-guide
  2. 02Vifu Serverresolve project + model
  3. 03Agent Gatewaymultiplexed invoke
  4. 04Agentlocal / self-hosted
Returnchat.completiontrace persisted

One endpoint, agents wherever they run

Use the AI client your game already understands.

Change the base URL, project key, and model slug. Vifu owns authorization, routing, cancellation, and trace persistence behind that boundary.

  1. 01

    Stable endpoint

    Games call the same API while the agent behind it changes.

  2. 02

    Project-scoped key

    One project key authorizes requests to the models bound inside that project.

  3. 03

    Agent Gateway

    One outbound connection carries many logical agent channels.

  4. 04

    Durable trace

    Every admitted request keeps its routing status and latency.

POST http://localhost:6790/my-project/v1/chat/completions
Authorization: Bearer vifu_pk_example
Content-Type: application/json

{
  "model": "town-guide",
  "messages": [
    { "role": "user", "content": "Open the north gate" }
  ],
  "stream": false
}

Operational evidence

See the endpoint behind the call.

The open-source Dashboard keeps projects, bindings, connection state, keys, and traces in one deployable system.

Vifu runtime console showing a game project and its connected agents
Runtime readyThe game calls the endpoint. Vifu records the route.

A stable boundary for changing agents

Design the game once. Evolve the agent separately.

AI NPCs

Keep a character integration stable while you replace or move its agent.

Agent playtests

Compare agent behavior without rebuilding the game-facing API.

Team prototypes

Share one reproducible endpoint contract across a changing project.

Playable demos

Run the agent boundary yourself instead of shipping provider credentials.

Self-hosted, end to end

Run the complete loop on your machine.

Start with the Apache-2.0 repository. Run the Runtime and Dashboard on infrastructure you control, connect your Agent Provider, and trace your game's requests.

  1. 01

    Prepare

    Copy .env.example to .env in the open-source Vifu checkout.

  2. 02

    Start

    Run docker compose up -d to launch PostgreSQL, Vifu, and the Dashboard.

  3. 03

    Connect

    Open localhost:6791, create the local administrator, connect OpenClaw, and expose an agent.

  4. 04

    Invoke

    Call localhost:6790/my-project/v1/chat/completions, then inspect the persisted trace.