JSPM

English  ·  简体中文  ·  Website

Agent Pipeline Orchestration Platform

Compose OpenClaw agents into executable pipelines — define, run, observe, and iterate. File-based persistence, zero external database.

[!TIP] TaskMeld is the pipeline runtime for OpenClaw. OpenClaw handles agent execution; TaskMeld chains agents into DAG workflows with routing, retries, and artifact tracking.

Prerequisites

• Node ≥ 18
• OpenClaw ≥ 5.20
• Windows (macOS and Linux are not yet tested)

Install

npm install -g taskmeld

Quick Start

# Initialize — guided setup for OpenClaw Gateway connection
taskmeld init

# Start the backend daemon
taskmeld server start

# List available pipelines
taskmeld pipeline list

# Run a pipeline
taskmeld pipeline start <pipelineId>

# Watch a pipeline run in real time
taskmeld pipeline watch <pipelineId>

Full command reference: taskmeld --help or CLI docs.

Features

• DAG Pipeline Engine — Node dependency graph, parallel groups, routing branches, per-node retry, state persistence
• CLI Tool — Full lifecycle: list, run, status, stop, retry, watch (WebSocket streaming)
• HTTP + WebSocket API — REST endpoints for control, WS broadcast for real-time observability
• Web Console — React 19 dashboard with DAG visualization, agent sessions, artifact browser, log viewer
• Gateway Integration — WebSocket client for OpenClaw Gateway auth, event relay, and agent session delegation
• File-based Persistence — All state stored as JSON and log files under ~/.taskmeld/ (TASKMELD_DATA_DIR can override it); zero external database

Architecture

CLI (taskmeld)  ·  Web Console (React)
        │                │
   HTTP API ─────── WS Broker
        │                │
     App Assembly (registry + runtime)
              │
     Pipeline Engine (DAG · scheduler · state machine)
              │
     Gateway Client (OpenClaw — auth, events, sessions)

Development Status

[!WARNING] TaskMeld is in its initial testing phase. Features are being built out incrementally, APIs may evolve between releases, and some surfaces are still rough. Production use is at your own discretion — we welcome early adopters and feedback.

Roadmap

Now

• Pipeline execution — Node-driven: each node binds an OpenClaw agent. The CLI exposes a full command set that external agents can invoke programmatically (pipeline list, pipeline start, pipeline status, etc.).
• Agent management — Primarily read-only (chat, edit core files like agent.md / memory.md / soul.md). Creating agents and configuring skills still requires switching to OpenClaw directly.
• Data storage — File-based persistence (JSON + log files under ~/.taskmeld/ by default), zero external dependencies.

Next

• Built-in autonomous agent — A first-class runtime component that owns the full pipeline lifecycle: scheduling runs, creating and reviewing pipeline definitions, triaging failures, and curating artifacts. The goal is to move from you operate the pipelines to the agent operates the pipelines, and you steer.
• Agent lifecycle management — Agent creation, skill configuration, and core file editing unified within TaskMeld, driven by the built-in agent — no more switching to OpenClaw.
• Database-backed storage — Evolve from file-based persistence to a database storage layer, improving query performance, concurrent access, and scalability while keeping a zero-dependency path for single-node setups.

Development

npm install          # Install dependencies
npm run build        # Build
npm run typecheck    # Type check only
npm run lint         # Lint
npm test             # Run tests
npm run dev:web      # Start frontend dev server (Vite HMR)

Documentation

• CLI Reference
• Backend Architecture
• Frontend Architecture
• Pipeline Engine
• Contributing

_{MIT — see LICENSE}