@d3ara1n/pi-provider-agnes

Agnes AI provider for pi — registers agnes (token billing) and agnes-plan (subscription) providers with text and image input models

Packages

Package details

extension

Install @d3ara1n/pi-provider-agnes from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:@d3ara1n/pi-provider-agnes

Package: @d3ara1n/pi-provider-agnes
Version: 1.3.1
Published: Oct 11, 2026
Downloads: 459/mo · 394/wk
Author: d3ara1n
License: MIT
Types: extension
Size: 25.4 KB
Dependencies: 0 dependencies · 1 peer

Pi manifest JSON

{
  "extensions": [
    "./src/index.ts"
  ]
}

Security note

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

README

@d3ara1n/pi-provider-agnes

Agnes AI provider for pi — registers two providers sharing the same text + image models but differing in billing model.

Providers

Provider ID	Name	Billing	API Key Env
`agnes`	Agnes AI	Token billing	`$AGNES_API_KEY`
`agnes-plan`	Agnes AI (Token Plan)	Subscription plan	`$AGNES_PLAN_API_KEY`

`agnes`

Token-based billing provider. Agnes has not published official token pricing, so pi shows no cost estimates for these models.

`agnes-plan`

Subscription plan provider. Cost set to zero — the subscription fee is a fixed monthly charge, not per-token.

Models

Both providers export the same chat model list, refreshed from the live API (GET /v1/models) when the cached snapshot is stale — at most once every 4 hours, matching pi's built-in catalog cadence. pi update --models forces an immediate refresh; the last-seen catalog is restored on offline startups. The endpoint returns ids only, so context windows, modalities and thinking behavior fall back to the shipped specs per model id; models new to the catalog get family defaults (text+image input, Qwen-style thinking toggle). Image-generation (agnes-image-*) and video-generation (agnes-video-*) models are excluded automatically.

Shipped chat specs (verified against the live catalog 2026-10-10):

Model	Reasoning	Input	Context	Max Output
`agnes-2.5-flash`	Yes	text, image	512K	65.5K
`agnes-2.0-flash`	Yes	text, image	256K	64K
`agnes-2.5-pro` / `-beta` / `-alpha`	Yes	text, image	256K*	64K*
`agnes-3.0-flash` / `-flash-max`	Yes	text, image	256K*	64K*

* endpoint publishes no metadata — family defaults, pending a live check. agnes-1.5-flash is no longer listed by the live catalog and was removed (2026-10); it re-registers automatically if it returns.

pi install npm:@d3ara1n/pi-provider-agnes

Or add to ~/.pi/agent/settings.json:

{
  "extensions": [
    "/absolute/path/to/pi-extensions/packages/pi-provider-agnes"
  ]
}

Set your API key(s) via environment variable:

export AGNES_API_KEY="sk-..."       # for agnes provider
export AGNES_PLAN_API_KEY="sk-..."  # for agnes-plan provider

Or configure via /login or auth.json.

Dependencies

None — this is a standalone provider with no pi-extension dependencies. It uses pi's built-in openai-completions streaming.

Usage Quota Reporting

Not yet implemented — Agnes AI does not currently expose a public quota or balance API. When one becomes available, quota reporting will be added via @d3ara1n/pi-usage-block-core.