@stigmergy/openclaw-scheduler

Zero-code pressure-field scheduling for OpenClaw. Install the plugin, add config, restart. Your agents get task scheduling with dependencies, priorities, and automatic retry. No code changes required.

Scheduling overhead is sub-millisecond regardless of agent count. Scales to thousands of tasks without configuration changes -- the bottleneck is your LLM provider, not the scheduler.

Install

Step 1 — Clone this repo and install the plugin:

git clone https://github.com/Production-Grade/stigmergy.git
openclaw plugins install ./stigmergy/packages/openclaw

Step 2 — Add this to your openclaw.json:

{
  "plugins": {
    "entries": {
      "stigmergy-scheduler": {
        "enabled": true,
        "config": {
          "tickIntervalMs": 5000,
          "defaultWakeThreshold": 0.4,
          "decayHalfLifeSeconds": 300,
          "evaporationThreshold": 0.01
        }
      }
    }
  }
}

Not sure where your config file is? Run openclaw config path to find it.

Step 3 — Restart OpenClaw.

You should see this in the logs:

Stigmergy scheduler started (tick=5000ms, threshold=0.4)

Run openclaw plugins list to confirm the plugin shows as loaded.

What you get

4 agent tools:

| Tool | Description | |------|-------------| | stigmergy_add_task | Queue a task with dependencies and priority. Dispatched when pressure exceeds threshold. | | stigmergy_inject_urgent | Force-inject a high-priority task for immediate processing. | | stigmergy_status | View scheduler metrics — pressure per agent, pending tasks, completions. | | stigmergy_deposit_signal | Manually deposit a pressure signal (e.g., from an external event). |

Hooks:

• agent_end — deposits activity signals when any agent run completes, creating natural pressure accumulation.

Gateway methods:

• stigmergy.metrics — full scheduler metrics snapshot
• stigmergy.pressure — per-agent pressure map and signal count

Config options

| Option | Default | Description | |--------|---------|-------------| | tickIntervalMs | 5000 | How often the scheduler evaluates and dispatches (ms) | | defaultWakeThreshold | 0.4 | Minimum pressure to wake an agent (0.0–1.0) | | decayHalfLifeSeconds | 300 | Half-life for signal decay | | evaporationThreshold | 0.01 | Signals below this are removed |

How it works

Instead of cron-based scheduling, agents wake when accumulated pressure from deposited signals exceeds a threshold. Signals decay exponentially over time (like pheromones). Task completions deposit new signals that trigger dependent work.

See the main README for the full explanation and benchmarks.

License

MIT