OpenClaw integration

Entroly is a context assurance engine for OpenClaw

Use Entroly inside OpenClaw to assemble budget-aware context, preserve provider-independent receipts, recover exact omitted messages, and optionally verify a draft before delivery—without replacing OpenClaw's Gateway, transcript, provider routing, or authentication.

Direct answer: Entroly adds a first-class context engine to OpenClaw. OpenClaw remains the agent runtime; Entroly controls the evidence sent into each model call and records the resulting context decisions.

Install Entroly for OpenClaw

The npm package is the currently verifiable public install source:

pip install -U entroly
openclaw plugins install npm:entroly-openclaw
openclaw plugins enable entroly
openclaw gateway restart

ClawHub is supported only when openclaw plugins search "entroly" returns a current public listing and install hint. Do not guess a ClawHub package name.

Select the context engine in ~/.openclaw/openclaw.json:

{
  plugins: {
    slots: {
      contextEngine: "entroly"
    }
  }
}

Then verify the loaded runtime:

openclaw plugins inspect entroly --runtime --json
openclaw plugins doctor

What Entroly adds to OpenClaw

Budget-aware evidence

Entroly scores older normalized messages against the current request, reserves budget for matching evidence, and keeps recent or protected content verbatim.

Context Receipts

Receipts record message-level decisions, estimated tokens, warnings, transformation state, and cryptographic commitments without persisting the request text or matched terms.

Exact recovery

Optional proof-guided recovery can identify omitted supporting messages and return exact text with SHA-256 commitments through a bounded revision request.

Fail-open safety

If a trusted model budget cannot be resolved or the bridge fails, Entroly returns the exact original OpenClaw context instead of risking a malformed or over-compressed request.

OpenClaw stays authoritative

OpenClaw owns the persisted transcript, provider credentials, routing, billing behavior, retries, and delivery.
System messages, tool calls, images, signatures, reasoning blocks, and unknown content blocks are not rewritten.
Entroly does not claim provider savings unless provider-bound usage is actually observable.
OpenClaw's resolved token budget wins over Entroly fallback metadata.

Frequently asked questions

Is Entroly an OpenClaw skill?

No. Entroly integrates as a context-engine plugin, which lets it operate on OpenClaw's normalized messages before a model run.

Does Entroly work with every OpenClaw provider?

The context engine operates before provider dispatch, so the same assembly path can support OpenClaw providers while OpenClaw remains responsible for provider-specific authentication and wire formats.

Does proof-guided recovery make another model call?

Only when explicitly enabled. Entroly returns a bounded revision request to OpenClaw; OpenClaw performs the provider call using its normal routing and billing path.

Where are receipts stored?

By default they are written under the workspace's .entroly/receipts/openclaw/ directory, with configurable quotas and local permissions.

Inspect the OpenClaw integration source · Compare all Entroly agent integrations