pi-model-request

Issue a one-off model request from a pi extension the same way Pi core does: resolve auth, add provider session headers, call the completion

Packages

Package details

extension

Install pi-model-request from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:pi-model-request

Package: pi-model-request
Version: 0.1.1
Published: Oct 4, 2026
Downloads: 498/mo · 15/wk
Author: maplezzk
License: MIT
Types: extension
Size: 14.4 KB
Dependencies: 0 dependencies · 1 peer

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

pi-model-request

Issue a one-off model request from a pi extension the same way Pi core does.

Pi sends its own model requests through the core streamFn, which resolves auth and injects provider attribution headers (core/model-runtime.js, core/provider-attribution.js in @earendil-works/pi-coding-agent). An extension that calls completeSimple / complete from @earendil-works/pi-ai has to do both steps itself, and missing one breaks the request: without the session header, opencode / opencode-go answers

400 MissingSessionID: Request is missing x-opencode-session

This package collapses "make a model request from an extension, the way Pi does" into one place: resolve auth, add the provider session headers, override the model base URL when auth resolves one, then call the completion you supply.

中文文档

Install

npm install pi-model-request

Quick start

import { completeSimple } from "@earendil-works/pi-ai/compat";
import { createModelRequester } from "pi-model-request";

// ctx: { modelRegistry, sessionManager } — an ExtensionContext satisfies this.
const request = createModelRequester(ctx, {
  base: completeSimple, // or `complete`
  authError: (error) => new Error(`Summarizer authentication failed: ${error}`),
});

const response = await request(model, context, { maxTokens: 2048, signal });

API

Export	Description
`createModelRequester(ctx, options?)`	Builds a completion with the same signature as the one you pass in `base` (default `completeSimple`). It resolves auth, merges the provider session headers, applies a resolved `baseUrl`, and throws on auth failure.
`resolveModelRequestAuth(ctx, model)`	The auth + header step on its own, as a `{ ok: true, ... } \| { ok: false, error }` result. Use it when the caller needs its own error handling (for example an injected auth resolver).
`mergeProviderSessionHeaders(model, sessionId, headers?)`	Merges the session headers into existing headers. Caller-supplied headers win, matching core merge order.
`getProviderSessionHeaders(model, sessionId)`	Builds `{ "x-opencode-session": sessionId, "x-opencode-client": "pi" }`, or `undefined` when not needed.
`requiresProviderSessionHeader(model)`	Whether the model belongs to the opencode family (provider id `opencode` / `opencode-go`, or an `opencode.ai` base URL host).
`ModelRequestAuthError`	Thrown by `createModelRequester` when no `authError` mapper is given; the provider error text is kept in `providerError`.

Scope

Only the opencode session headers are replicated. OpenRouter / NVIDIA / Cloudflare attribution headers are gated by Pi's install-telemetry setting, which an extension cannot read, so this package deliberately does not guess.
Values must stay identical to core (x-opencode-client: pi). When core changes its rules, update this package in the same change.
This affects request routing and attribution only. It is not an authentication mechanism and grants no extra access.

License

MIT