pi-openai-codex-fast
Pi package that adds an openai-codex-fast provider backed by openai-codex with priority service tier
Package details
Install pi-openai-codex-fast from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:pi-openai-codex-fast- Package
pi-openai-codex-fast- Version
0.0.10- Published
- Jul 17, 2026
- Downloads
- 1,702/mo · 542/wk
- Author
- kaanozdokmeci
- License
- MIT
- Types
- extension
- Size
- 16 KB
- Dependencies
- 0 dependencies · 2 peers
Pi manifest JSON
{
"extensions": [
"./index.ts"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
pi-openai-codex-fast
Pi package that adds an openai-codex-fast provider backed by built-in openai-codex with serviceTier: "priority".
Behavior
openai-codex-fast is a separate selectable provider that delegates to Pi's built-in openai-codex implementation with the same model id and serviceTier: "priority". Normal openai-codex/<modelId> selections are left on the normal/default-tier path.
Currently exposed fast models:
gpt-5.6-lunagpt-5.6-terragpt-5.6-solgpt-5.5gpt-5.4gpt-5.4-mini
Runtime behavior when openai-codex-fast/<modelId> is selected:
- Reuses existing
openai-codexauth from Pi auth storage. - Sends Codex requests through the built-in Codex response API with
serviceTier: "priority". - Stores all generated assistant messages canonically as built-in Codex, including normal replies, tool-calling replies, and setup/error/aborted replies:
provider: "openai-codex"api: "openai-codex-responses"
- Does not rewrite stored assistant history back to
openai-codex-fastoropenai-codex-fast-responses.
Fast-mode recovery:
- No custom fast-mode session state is persisted.
- On any
session_startreason (startup,reload,new,resume, orfork), the extension scans the current branch backward for the latest overallmodel_change. - If that latest
model_changeisopenai-codex-fast/<modelId>, it selectsopenai-codex-fast/<modelId>again. - If the latest
model_changeis anything else, it does nothing and lets Pi's normal model recovery handle it. - The extension does not handle
session_tree, so branch switches do not trigger model reconciliation.
Install
Local path
pi install .
Temporary use
pi -e .
After install, log in to built-in Codex if needed:
/login openai-codex
Then select a fast model with /model, for example:
openai-codex-fast/gpt-5.5
Local development
npm install
npm test # typecheck + integration tests for both direct TS loading and built JS
npm run check
npm run build
npm run test:ts
npm run test:js
npm run lint
npm run fmt
hk check --all --check
npm run benchmark
Packaging
This package publishes the TypeScript extension entrypoint and these project files explicitly:
index.tsREADME.mdCHANGELOG.mdLICENSE
The build output is a local test artifact for verifying the extension also works as native JavaScript; it is not published.
Release flow:
- Create a release commit named
release: vX.Y.Z. - Tag that commit as
vX.Y.Z. - Push
mainand the tag to GitHub. - The tag push triggers GitHub Actions to stage the package on npm via trusted publishing with npm provenance.
- Approve the staged package on npmjs.com, or with
npm stage approve <stage-id>.
Prerelease tags such as vX.Y.Z-alpha.N use the same CI flow. CI derives the npm dist-tag from the first prerelease identifier (alpha for X.Y.Z-alpha.N, beta for X.Y.Z-beta.N, and so on); stable versions use latest.