Daeva

Local GPU pod orchestrator for AI workloads. One process to register, schedule, and manage containerized AI services (pods) on a single host with shared GPU resources.

Features

• Pod registry — register and discover GPU-backed AI services
• Job queue — submit async jobs, route them to capable pods, track results
• Exclusivity groups — automatic stop/switch when pods contend for the same GPU
• Pod packages — portable pod-package.json format for distributing pod definitions
• Package install — materialize packages from local paths, Git repos, archives, or registries
• MCP server — expose orchestrator capabilities to AI coding assistants via daeva-mcp
• Built-in pods — bundled compatibility manifests for ComfyUI, Whisper, and OCR/Vision

Install

npm install -g @asmostans/daeva

Or from source:

git clone https://github.com/openclaw/daeva.git
cd daeva
npm install
npm run build

Platform installers

Platform-specific install scripts handle Node.js, Podman, and service setup:

# Linux (apt/dnf/pacman)
curl -fsSL https://asmo.bot/install-linux.sh | bash

# macOS (Homebrew)
curl -fsSL https://asmo.bot/install-macos.sh | bash

# Windows (PowerShell, winget/choco)
irm https://asmo.bot/install-windows.ps1 | iex

All scripts support --dry-run, --skip-podman, --skip-service, and more. Run with --help for details.

Quickstart

Start the server:

daeva
# Listening on http://127.0.0.1:8787

Submit a job:

curl -X POST http://127.0.0.1:8787/jobs \
  -H 'Content-Type: application/json' \
  -d '{"type": "transcribe-audio", "files": [{"source": "path", "path": "/tmp/demo.wav"}]}'

Check status:

curl http://127.0.0.1:8787/status

List pods:

curl http://127.0.0.1:8787/pods

Install a pod package:

curl -X POST http://127.0.0.1:8787/pods/create \
  -H 'Content-Type: application/json' \
  -d '{"alias": "comfyapi"}'

Then point Comfy clients at Daeva's proxy instead of raw port 8188:

export DAEVA_BASE=http://127.0.0.1:8787
curl "$DAEVA_BASE/proxy/comfyapi/system_stats"

MCP server

Daeva ships a stdio-based MCP server binary for integration with AI coding assistants:

daeva-mcp --base-url http://127.0.0.1:8787

Add to your MCP client config:

{
  "command": "daeva-mcp",
  "args": ["--base-url", "http://127.0.0.1:8787"]
}

Configuration

REST API

Core endpoints

Observability

Pod packages

Daeva uses a portable pod-package.json format. Minimal example:

{
  "schemaVersion": "1",
  "packageType": "pod-package",
  "name": "daeva-whisper",
  "version": "0.1.0",
  "pod": {
    "id": "whisper",
    "nickname": "Whisper",
    "description": "Speech-to-text pod",
    "manifestVersion": "1",
    "capabilities": ["speech-to-text"],
    "runtime": {
      "kind": "http-service",
      "baseUrl": "http://127.0.0.1:8001",
      "submitPath": "/transcribe",
      "method": "POST"
    }
  }
}

See examples/whisper-pod-package/ for a complete example with Dockerfile, deploy scripts, and directory scaffolding.

Package sources

Packages can be installed from:

• local-file — local directory with a pod-package.json
• github-repo — GitHub owner/repo with optional ref and subpath
• git-repo — arbitrary Git URL
• uploaded-archive — .tar.gz or .zip uploaded directly
• registry-index — delegated lookup from a registry catalog

During install, Daeva now runs package install hooks, creates declared host directories, and persists resolved host-path template variables (for example MODELS_DIR, INPUT_DIR, HOST_DIR_1) alongside installed package metadata.

Built-in pods

Project layout

src/server.ts              Fastify app and routes
src/cli.ts                 CLI entry point (daeva binary)
src/mcp-cli.ts             MCP server entry point (daeva-mcp binary)
src/job-manager.ts         Queue + job lifecycle
src/router.ts              Job-to-pod routing
src/pod-controller.ts      Runtime state + exclusivity switching
src/registry.ts            Pod registry + alias resolution
src/adapters.ts            Execution adapter abstraction
src/schemas.ts             Zod schemas for manifests + validation
src/manifests/builtin.ts   Bundled pod definitions
examples/                  Pod package examples
scripts/                   Platform install scripts

Development

npm install
npm run dev          # dev server with tsx
npm test             # vitest
npm run typecheck    # tsc --noEmit
npm run build        # compile to dist/

License

MIT