Codex routing and deployment

Which effective routes select the Codex runtime, and the deployment shapes built on that policy. Part of the Codex harness guide; Where each section moved lists every section.

Routing and model selection

openai/gpt-6-astra defaults to medium reasoning effort through the shared OpenAI provider policy when the account supports it. For OpenClaw-managed turns, the resolved effort is sent in Codex turn/start requests, including collaborationMode.settings.reasoning_effort, so the native thread uses the same default as Control UI. Explicit thinking settings still win; an existing session or agent configured for high stays at high. Threads attached with native settings preserved retain their native effort.

Keep provider refs and runtime policy separate:

• Use openai/gpt-* for canonical OpenAI model selection. The prefix alone never selects Codex.
• With runtime unset or auto, only an exact official HTTPS Platform Responses or ChatGPT Responses route with no authored provider request override may select Codex implicitly. Valid model-scoped Fast-mode and cutoff controls do not count as authored request params.
• Do not use legacy Codex GPT refs in config; run openclaw doctor --fix to repair legacy refs and stale session route pins.
• agentRuntime.id: "codex" makes Codex a fail-closed requirement for a compatible route. It does not make an incompatible effective route compatible.
• agentRuntime.id: "openclaw" opts a provider or model into the embedded OpenClaw runtime when that is intentional.
• /codex ... controls native Codex app-server conversations from chat.
• ACP/acpx is a separate external harness path. Use it only when the user asks for ACP/acpx or an external harness adapter.

User intent	Use
Attach the current chat	/codex bind [thread-id] [--cwd <path>] [--model <model>] [--provider <provider>]
Resume an existing Codex thread	/codex resume <thread-id>
List or filter Codex threads	/codex threads [filter]
Read or update the bound thread’s native goal	/codex goal [status|set <objective>|pause|resume|block|complete|clear]
List native Codex plugins	/codex plugins list
Discover available native Codex marketplace plugins	/codex plugins available
Install and authorize one native Codex plugin	/codex plugins install <plugin>@<marketplace>
Enable or disable a configured native Codex plugin	/codex plugins enable <name>, /codex plugins disable <name>
Resume a stored Codex CLI session as a paired-node turn	/codex sessions --host <node> [filter], then /codex resume <session-id> --host <node> --bind here
View non-archived Codex sessions across computers	Enable Codex supervision and open Codex Sessions
Change the bound thread’s model, fast-mode, or permissions	/codex model <model>, /codex fast [on|off|status], /codex permissions [default|yolo|status]
Compact the current Codex session	/codex compact
Stop or steer the active turn	/codex stop, /codex steer <text>
Detach the current binding	/codex detach (alias /codex unbind)
Send Codex feedback only	/codex diagnostics [note]
Start an ACP/acpx task	ACP/acpx session commands, not /codex

Use case	Configure	Verify	Notes
Eligible OpenAI route with native Codex runtime	Exact official HTTPS Responses/ChatGPT route with no authored provider request override, plus enabled codex plugin	/status shows Runtime: OpenAI Codex	Valid Fast runtime controls do not disqualify this path
Fail closed if Codex is unavailable	Provider or model agentRuntime.id: "codex"	Missing harness fails the turn	Authored request overrides may still use declared fallback
Direct OpenAI API-key traffic through OpenClaw	Provider or model agentRuntime.id: "openclaw" and normal OpenAI auth	/status shows OpenClaw runtime	Use only when OpenClaw is intentional
Legacy config	legacy Codex GPT refs	openclaw doctor --fix rewrites it	Do not write new config this way
ACP/acpx Codex adapter	ACP sessions_spawn({ runtime: "acp" })	ACP task/session status	Separate from native Codex harness

agents.defaults.imageModel follows the same prefix split. Use openai/gpt-* for the normal OpenAI route and codex/gpt-* only when image understanding should run through a bounded Codex app-server turn. Doctor rewrites legacy Codex GPT refs to openai/gpt-*.

Operator role model permissions

A role’s model policy applies to each native inference request, including retries, child turns, ordinary reviews, native image and search requests, and compaction. OpenClaw checks the actual model against both the work’s original permissions and current policy before forwarding it. Verified catalog-to-native model mappings remain valid. Removing a model cancels affected inference while permitted work continues.

Reusing a native child does not attach its creator’s permissions permanently. A later turn with unambiguous source attribution uses its submitting operator’s authority after the earlier turn settles. With the native hook integration active, sending input to an active turn requires the same original authority, including follow-up requests that native execution can consume in that turn. Native Guardian reviews and configured memory processing keep their existing service authority, verified from native request provenance and their owning execution or configured service.

Restricted runs require an OpenClaw-owned native inference route. Managed stdio connections can retain HTTP or WebSocket Responses providers over public HTTPS; already-owned native bindings can reuse that route. Provider projection shares the connection’s eight-route limit; excess providers remain unavailable to restricted native restores, and selecting a new route after capacity is reached requires a fresh managed native connection. Custom providers need an explicit native base_url; query fields use the native query_params table. Unowned attachments, native local-model providers, AWS-signed requests, system-proxy profiles, custom native certificate files, and proxy settings that cannot preserve both upstream routing and private loopback access cannot establish this guarantee. OpenClaw rejects a restricted run on those paths before starting it. Roles without a model policy keep their existing native connection and optional-hook behavior. Introducing a model policy while an unqualified operator execution is active cancels directly owned and otherwise unambiguously bound work, including its unqualified children.

When staff run without a model policy and qualified native hooks are disabled or unavailable, native execution can mix accepted input into an existing turn or combine queued input from several senders into a new turn without preserving unique sender attribution. Previously accepted unrestricted input that can no longer be attributed uniquely may continue under the receiving execution’s valid authority after a contributing sender’s authorization ends or becomes restricted. A matching native root alone does not prove unique attribution. OpenClaw does not interrupt independently authorized receiver work to guess which input it consumed.

This limitation does not exempt newly restricted work or revocation of a directly or otherwise unambiguously bound source. Exact attribution requires the qualified native hook integration. Visitor Access requires an explicit model policy; its Codex runs require the qualified integration, so its normal restricted flow cannot enter this optional staff configuration.

Deployment patterns

Basic Codex deployment

Use the quickstart config for an OpenAI model whose effective official HTTPS route is eligible to select Codex implicitly:

{
  plugins: {
    entries: {
      codex: {
        enabled: true,
      },
    },
  },
  agents: {
    defaults: {
      model: "openai/gpt-6-astra",
    },
  },
}

Mixed provider deployment

Configure a Claude main agent and add a named Codex agent:

{
  plugins: {
    entries: {
      codex: {
        enabled: true,
      },
    },
  },
  agents: {
    ownership: "explicit",
    defaults: {
      model: "anthropic/claude-opus-4-6",
    },
    entries: {
      main: {
        model: "anthropic/claude-opus-4-6",
      },
      codex: {
        name: "Codex",
        model: "openai/gpt-6-astra",
      },
    },
  },
}

This explicit fleet has no default agent; target main or codex with a session, --agent, or binding. The main agent uses its normal provider path. The codex agent uses Codex app-server when its effective OpenAI route remains compatible; add explicit model-scoped agentRuntime.id: "codex" when that should be a fail-closed requirement.

Fail-closed Codex deployment

An eligible exact official HTTPS OpenAI route can resolve to Codex when the bundled plugin is available. Add explicit runtime policy for a written fail-closed rule:

{
  models: {
    providers: {
      openai: {
        agentRuntime: {
          id: "codex",
        },
      },
    },
  },
  agents: {
    defaults: {
      model: "openai/gpt-6-astra",
    },
  },
  plugins: {
    entries: {
      codex: {
        enabled: true,
      },
    },
  },
}

With Codex forced, OpenClaw fails early if the plugin is disabled, the app-server is too old or cannot start, or route/auth support is rejected without a declared fallback. Authored request overrides may instead use the selection-time OpenClaw fallback that preserves the exact request. Once Codex starts, its failures are not replayed through OpenClaw.