vultr/openclaw
1
0
Fork 0
mirror of https://github.com/openclaw/openclaw.git synced 2026-04-12 01:31:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs: add codex harness setup guide

Browse Source
This commit is contained in:
Peter Steinberger
2026-04-10 21:45:23 +01:00
parent b76f218c53
commit c66afe472a
4 changed files with 303 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/concepts/model-providers.md
View File

@@ -56,7 +56,7 @@ For model selection rules, see [/concepts/models](/concepts/models).
to use the OpenAI provider and the normal OpenClaw provider transport.
Codex-only deployments can disable automatic PI fallback with
`agents.defaults.embeddedHarness.fallback: "none"`; see
[Agent Harness Plugins](/plugins/sdk-agent-harness#disable-pi-fallback).
[Codex Harness](/plugins/codex-harness).
## Plugin-owned provider behavior

1
docs/docs.json
View File

@@ -1113,6 +1113,7 @@
"tools/plugin",
"plugins/community",
"plugins/bundles",
"plugins/codex-harness",
"plugins/webhooks",
"plugins/voice-call",
{

297
docs/plugins/codex-harness.md Normal file
View File

@@ -0,0 +1,297 @@
---
title: "Codex Harness"
summary: "Run OpenClaw embedded agent turns through the bundled Codex app-server harness"
read_when:
- You want to use the bundled Codex app-server harness
- You need Codex model refs and config examples
- You want to disable PI fallback for Codex-only deployments
---
# Codex Harness
The bundled `codex` plugin lets OpenClaw run embedded agent turns through the
Codex app-server instead of the built-in PI harness.
Use this when you want Codex to own the low-level agent session: model
discovery, native thread resume, native compaction, and app-server execution.
OpenClaw still owns chat channels, session files, model selection, tools,
approvals, media delivery, and the visible transcript mirror.
The harness is off by default. It is selected only when the `codex` plugin is
enabled and the resolved model is a `codex/*` model, or when you explicitly
force `embeddedHarness.runtime: "codex"` or `OPENCLAW_AGENT_RUNTIME=codex`.
If you never configure `codex/*`, existing PI, OpenAI, Anthropic, Gemini, local,
and custom-provider runs keep their current behavior.
## Pick the right model prefix
OpenClaw has separate routes for OpenAI and Codex-shaped access:
| Model ref | Runtime path | Use when |
| ---------------------- | -------------------------------------------- | ----------------------------------------------------------------------- |
| `openai/gpt-5.4` | OpenAI provider through OpenClaw/PI plumbing | You want direct OpenAI Platform API access with `OPENAI_API_KEY`. |
| `openai-codex/gpt-5.4` | OpenAI Codex OAuth provider through PI | You want ChatGPT/Codex OAuth without the Codex app-server harness. |
| `codex/gpt-5.4` | Bundled Codex provider plus Codex harness | You want native Codex app-server execution for the embedded agent turn. |
The Codex harness only claims `codex/*` model refs. Existing `openai/*`,
`openai-codex/*`, Anthropic, Gemini, xAI, local, and custom provider refs keep
their normal paths.
## Requirements
- OpenClaw with the bundled `codex` plugin available.
- Codex app-server `0.118.0` or newer.
- Codex auth available to the app-server process.
The plugin blocks older or unversioned app-server handshakes. That keeps
OpenClaw on the protocol surface it has been tested against.
For live and Docker smoke tests, auth usually comes from `OPENAI_API_KEY`, plus
optional Codex CLI files such as `~/.codex/auth.json` and
`~/.codex/config.toml`. Use the same auth material your local Codex app-server
uses.
## Minimal config
Use `codex/gpt-5.4`, enable the bundled plugin, and force the `codex` harness:
```json5
{
plugins: {
entries: {
codex: {
enabled: true,
},
},
},
agents: {
defaults: {
model: "codex/gpt-5.4",
embeddedHarness: {
runtime: "codex",
fallback: "none",
},
},
},
}
```
If your config uses `plugins.allow`, include `codex` there too:
```json5
{
plugins: {
allow: ["codex"],
entries: {
codex: {
enabled: true,
},
},
},
}
```
Setting `agents.defaults.model` or an agent model to `codex/<model>` also
auto-enables the bundled `codex` plugin. The explicit plugin entry is still
useful in shared configs because it makes the deployment intent obvious.
## Add Codex without replacing other models
Keep `runtime: "auto"` when you want Codex for `codex/*` models and PI for
everything else:
```json5
{
plugins: {
entries: {
codex: {
enabled: true,
},
},
},
agents: {
defaults: {
model: {
primary: "codex/gpt-5.4",
fallbacks: ["openai/gpt-5.4", "anthropic/claude-opus-4-6"],
},
models: {
"codex/gpt-5.4": { alias: "codex" },
"codex/gpt-5.4-mini": { alias: "codex-mini" },
"openai/gpt-5.4": { alias: "gpt" },
"anthropic/claude-opus-4-6": { alias: "opus" },
},
embeddedHarness: {
runtime: "auto",
fallback: "pi",
},
},
},
}
```
With this shape:
- `/model codex` or `/model codex/gpt-5.4` uses the Codex app-server harness.
- `/model gpt` or `/model openai/gpt-5.4` uses the OpenAI provider path.
- `/model opus` uses the Anthropic provider path.
- If a non-Codex model is selected, PI remains the compatibility harness.
## Codex-only deployments
Disable PI fallback when you need to prove that every embedded agent turn uses
the Codex harness:
```json5
{
agents: {
defaults: {
model: "codex/gpt-5.4",
embeddedHarness: {
runtime: "codex",
fallback: "none",
},
},
},
}
```
Environment override:
```bash
OPENCLAW_AGENT_RUNTIME=codex \
OPENCLAW_AGENT_HARNESS_FALLBACK=none \
openclaw gateway run
```
With fallback disabled, OpenClaw fails early if the Codex plugin is disabled,
the requested model is not a `codex/*` ref, the app-server is too old, or the
app-server cannot start.
## Per-agent Codex
You can make one agent Codex-only while the default agent keeps normal
auto-selection:
```json5
{
agents: {
defaults: {
embeddedHarness: {
runtime: "auto",
fallback: "pi",
},
},
list: [
{
id: "main",
default: true,
model: "anthropic/claude-opus-4-6",
},
{
id: "codex",
name: "Codex",
model: "codex/gpt-5.4",
embeddedHarness: {
runtime: "codex",
fallback: "none",
},
},
],
},
}
```
Use normal session commands to switch agents and models. `/new` creates a fresh
OpenClaw session and the Codex harness creates or resumes its sidecar app-server
thread as needed. `/reset` clears the OpenClaw session binding for that thread.
## Model discovery
By default, the Codex plugin asks the app-server for available models. If
discovery fails or times out, it uses the bundled fallback catalog:
- `codex/gpt-5.4`
- `codex/gpt-5.4-mini`
- `codex/gpt-5.2`
You can tune discovery under `plugins.entries.codex.config.discovery`:
```json5
{
plugins: {
entries: {
codex: {
enabled: true,
config: {
discovery: {
enabled: true,
timeoutMs: 2500,
},
},
},
},
},
}
```
Disable discovery when you want startup to avoid probing Codex and stick to the
fallback catalog:
```json5
{
plugins: {
entries: {
codex: {
enabled: true,
config: {
discovery: {
enabled: false,
},
},
},
},
},
}
```
## Tools, media, and compaction
The Codex harness changes the low-level embedded agent executor only.
OpenClaw still builds the tool list and receives dynamic tool results from the
harness. Text, images, video, music, TTS, approvals, and messaging-tool output
continue through the normal OpenClaw delivery path.
When the selected model uses the Codex harness, native thread compaction is
delegated to Codex app-server. OpenClaw keeps a transcript mirror for channel
history, search, `/new`, `/reset`, and future model or harness switching.
Media generation does not require PI. Image, video, music, PDF, TTS, and media
understanding continue to use the matching provider/model settings such as
`agents.defaults.imageGenerationModel`, `videoGenerationModel`, `pdfModel`, and
`messages.tts`.
## Troubleshooting
**Codex does not appear in `/model`:** enable `plugins.entries.codex.enabled`,
set a `codex/*` model ref, or check whether `plugins.allow` excludes `codex`.
**OpenClaw falls back to PI:** set `embeddedHarness.fallback: "none"` or
`OPENCLAW_AGENT_HARNESS_FALLBACK=none` while testing.
**The app-server is rejected:** upgrade Codex so the app-server handshake
reports version `0.118.0` or newer.
**Model discovery is slow:** lower `plugins.entries.codex.config.discovery.timeoutMs`
or disable discovery.
**A non-Codex model uses PI:** that is expected. The Codex harness only claims
`codex/*` model refs.
## Related
- [Agent Harness Plugins](/plugins/sdk-agent-harness)
- [Model Providers](/concepts/model-providers)
- [Configuration Reference](/gateway/configuration-reference)
- [Testing](/help/testing#live-codex-app-server-harness-smoke)

4
docs/plugins/sdk-agent-harness.md
View File

@@ -126,6 +126,9 @@ when you want Codex-managed auth, Codex model discovery, native threads, and
Codex app-server execution. `/model` can switch among the Codex models returned
by the Codex app server without requiring OpenAI provider credentials.
For operator setup, model prefix examples, and Codex-only configs, see
[Codex Harness](/plugins/codex-harness).
OpenClaw requires Codex app-server `0.118.0` or newer. The Codex plugin checks
the app-server initialize handshake and blocks older or unversioned servers so
OpenClaw only runs against the protocol surface it has been tested with.
@@ -257,4 +260,5 @@ on the same delivery path as PI-backed runs.
- [SDK Overview](/plugins/sdk-overview)
- [Runtime Helpers](/plugins/sdk-runtime)
- [Provider Plugins](/plugins/sdk-provider-plugins)
- [Codex Harness](/plugins/codex-harness)
- [Model Providers](/concepts/model-providers)