Code Mode quickstart

Enable code mode

The recommended path is Settings → Agents & Tools → Labs → Code Mode. The switch takes effect for future agent runs without restarting the Gateway and selects the "auto" tier. To enable the same tier without the Control UI, set it in config:

{
  tools: {
    codeMode: "auto",
  },
}

To default code mode on for every tool-capable run, regardless of model:

{
  tools: {
    codeMode: true,
  },
}

Object form works too: tools.codeMode.enabled accepts the same false, true, and "auto" values. Code mode stays off when tools.codeMode is omitted, false, or an object without an explicit enabled value, unless an agent or model override enables it. Configuring limits or other Code Mode options does not enable it. See Automatic per-model activation for the exact semantics and the shipped model list. If you use sandboxed agents with configured MCP servers, also allow the bundled MCP plugin in the sandbox tool policy, for example tools.sandbox.tools.alsoAllow: ["bundle-mcp"]. See Configuration - tools and custom providers. Set explicit limits for tighter bounds:

{
  tools: {
    codeMode: {
      enabled: true,
      timeoutMs: 10000,
      memoryLimitBytes: 67108864,
      maxOutputBytes: 65536,
      maxSnapshotBytes: 10485760,
      maxPendingToolCalls: 16,
      snapshotTtlSeconds: 900,
      searchDefaultLimit: 8,
      maxSearchLimit: 50,
    },
  },
}

Override one model

Set codeMode: true or codeMode: false on an exact provider/model entry in agents.defaults.models. Omit codeMode to inherit the parent activation setting, including its "auto" behavior. The model field accepts only a boolean; "auto" belongs on the global or per-agent tools.codeMode setting. Wildcard rows such as "openai/*" may configure runtime policy, but cannot set codeMode; config validation rejects them instead of ignoring the override.

{
  tools: { codeMode: "auto" },
  agents: {
    defaults: {
      models: {
        "openai/gpt-5.6-luna": {
          agentRuntime: { id: "openclaw" },
          codeMode: true,
        },
      },
    },
    entries: {
      research: {
        models: {
          "openai/gpt-5.6-luna": { codeMode: false },
        },
      },
    },
  },
}

The example enables Code Mode for this model except on the research agent. Activation resolves from the first explicit setting in this order:

agents.entries.<agent>.models["provider/model"].codeMode.
agents.entries.<agent>.tools.codeMode.enabled (or its boolean/"auto" shorthand).
agents.defaults.models["provider/model"].codeMode.
tools.codeMode.enabled (or its shorthand), defaulting to false.

In the Control UI, open Settings → Agents → Agent defaults, show Advanced settings, and find Models under Agent Defaults. Each model has a Code Mode selector beside its runtime: Default removes the override, On saves true, and Off saves false. For agent-specific overrides, expand Agent List, then the agent’s Agent Model Overrides. Unsupported fields remain marked for Raw editing without hiding the supported settings beside them. Overrides affect the selected model on future runs, including fallback models; they do not enable tools on a tool-free run or change runtime selection. The example separately selects agentRuntime.id: "openclaw" because OpenAI routes may otherwise use Codex. These settings do not control Codex native Code Mode. Model overrides change activation only; limits still come from the global and per-agent tools.codeMode options.

What the model does

For a tool with a declared output such as Array<{ id: string; paid: boolean; tons: number }>, one guest program can select, call, and transform it:

const [shipmentTool] = await catalog.search("list shipments");
const shipments = await shipmentTool({});
return shipments.filter((shipment) => !shipment.paid && shipment.tons > 10);

Declared output fields may feed later calls in that same exec; do not spend a second exec merely inspecting them. When a quick-index line ends in -> ?, the output shape is unknown. The first exec must return the final async tool call unchanged or save its value for a bounded preview. Do not feed the unknown value into guessed field-dependent logic in the same program. Inspect the raw value or saved preview, then use a later exec for dependent composition.

Reuse data across cells

Return an unfamiliar result normally. When a final object or array would exceed the output or model-result budget, interactive exec and wait automatically save it when capacity permits. The completed result contains value: { truncated: true, reference: { id, bytes, count, shape, preview, previewTruncated }, guidance }. Use that reference’s id with results.load(id) in another cell. Small returned values keep their ordinary shape; emitted text and json output is not automatically saved. Save a fetched result when later steps need to inspect and transform it:

const [shipmentTool] = await catalog.search("list shipments");
return await results.save(await shipmentTool({}));

Return this descriptor directly as the cell’s output. Its preview, shape, and count are already prepared; the full fetched value stays in saved storage. The returned reference contains an id, encoded JSON bytes, count, a sampled shape, and a bounded string preview. count is the top-level array length, object key count, or 1 for a scalar. Nested array lengths appear in the sampled view with paths and sampled item indices. The first, middle, and last items provide observed shapes, including observed heterogeneous values; these are not schemas or validation guarantees. Discovery visits at most 128 nodes, five levels, and 16 keys per object, prioritizing the eight largest arrays it finds. Limited traversal is labeled explicitly. Compact scalar envelope fields provide context alongside array samples. Small previews contain the complete JSON. Larger previews describe sampled data and may themselves be cut to a JSON prefix. previewTruncated: true always means the preview is incomplete. Descriptors fit within 768 encoded JSON bytes; output and model budgets may shorten their descriptive fields further while preserving their identity. Load the original JSON before processing full data. After inspecting those fields, a later cell can reuse the original result:

const shipments = await results.load("result_<id from the previous cell>");
return shipments.filter((shipment) => !shipment.paid).length;

Every load returns an independent JSON copy. Modifying it does not change the saved value. Use await results.delete(id) to release capacity. Missing or expired references reject with a catchable error. API.read("results.d.ts") provides TypeScript-style documentation; loaded data is declared as unknown, so inspect its shape before composing it. References last only for the current agent run and catalog. They survive cell completion and wait, but not run end, abort, catalog replacement, permission changes, or Gateway restart. They are snapshots: fetch again when current external state matters. The store holds at most 64 values with a total encoded JSON allowance of min(memoryLimitBytes, maxSnapshotBytes) (10 MiB by default), separate from the cell inbox. New saves fail when full; existing references are never evicted automatically. No functions, tool handles, or permissions are saved. Result operations are unavailable in restartSafe cells because references are transient and deletion cannot be replayed safely. Automatic preservation is also disabled in headless execution. If capacity, the data allowance, or the output budget prevents delivering a reference, the original call remains completed and returns an ordinary truncation marker explaining that the full result was not retained. Existing references remain unchanged; no partially saved value is advertised.

Recover from tool errors

Nested tool failures are ordinary JavaScript errors. Guest code can catch them and inspect diagnostic fields: code identifies input_contract, output_contract, invalid_contract, invalid_input, or tool_error; location contains the original guest call-site frames when available. effectStatus remains "unknown": classification is not a dispatch-owner receipt and never grants retry permission. In particular, a tool can throw an input error after starting work. Guest code can return the information needed to choose the next action:

try {
  return await terminal({ action: "list" });
} catch (error) {
  return { status: "unavailable", error: error.message };
}

For output_contract, the tool returned a response that failed its declared output schema. The error includes up to five validation details, each bounded to 256 UTF-8 bytes plus a truncation marker, with field paths and expected constraints. The response body is omitted. For example, receipt.count: must be number identifies a malformed count without printing the returned value. Check current state before retrying: result validation does not undo earlier effects, and effectStatus remains "unknown". Await every tool call or handle its rejection explicitly. OpenClaw drains dispatched calls before completing a cell; an unhandled rejection, including one from an unawaited call or timer callback, fails the cell instead of silently reporting success. Handlers attached after a suspension still handle their original promises. JavaScript syntax errors and uncaught nested tool failures become failed exec or wait results. The model can read the error, correct its code, inspect the current state, and continue with the normal tool surface. A failed cell does not impose a separate recovery mode or mutation budget. OpenClaw does not automatically replay a failed program. Earlier calls may have changed state, and a failed call may have partially applied. Inspect authoritative state before deciding what remains, and do not repeat completed actions. This also applies when wait resumes a suspended cell: its earlier calls belong to the same program. Every subsequent call runs the ordinary hooks and approvals again. Consumed voice confirmations stay consumed; continuing after an error does not restore a grant. Cancellation, explicitly terminal tool outcomes, sandbox restrictions, approval requirements, and tool-policy denials retain their existing behavior.

Verify the active surface

To confirm the model payload shape while debugging, run the Gateway with targeted logging:

OPENCLAW_DEBUG_CODE_MODE=1 \
OPENCLAW_DEBUG_MODEL_TRANSPORT=1 \
OPENCLAW_DEBUG_MODEL_PAYLOAD=tools \
openclaw gateway

With code mode active, the logged model-facing tool names should be exec and wait. For the full redacted provider payload, add OPENCLAW_DEBUG_MODEL_PAYLOAD=full-redacted for a short debugging session.

Use Swarm for agent fan-out

Swarm adds agents.run(), phase(), and log() guest globals for orchestrating concurrent sub-agents from Code Mode scripts. Swarm is enabled by default; Code Mode remains separately opt-in through "auto" or true. Use normal JavaScript control flow for fan-out, decision gates, and structured collection. The Swarm globals, API.read("agents.d.ts"), and Swarm prompt hints appear only when Swarm is enabled and the native OpenClaw sessions_spawn tool is present in the Code Mode catalog and permitted by the run’s execution allowlist. An MCP tool with the same name does not qualify. Code Mode waits for collector results internally, so agents.run() does not require the standalone agents_wait tool. Direct low-level Swarm use requires both tools allowed. Set tools.swarm: false or tools.swarm.enabled: false to opt out, globally or under an agent’s tools. Engaging Code Mode does not override that opt-out or grant access to tools denied by policy.

Overview

Plugins

Plugin guides

Building plugins

Skills

Automation

Tools

Agent coordination

Code Mode quickstart

Enable code mode

Override one model

What the model does

Reuse data across cells

Recover from tool errors

Verify the active surface

Use Swarm for agent fan-out

​Enable code mode

​Override one model

​What the model does

​Reuse data across cells

​Recover from tool errors

​Verify the active surface

​Use Swarm for agent fan-out

Enable code mode

Override one model

What the model does

Reuse data across cells

Recover from tool errors

Verify the active surface

Use Swarm for agent fan-out