pi-our-free-model1.0.2

Free frontier models (DeepSeek V4.1 Flash, Kimi K3, GLM-5.3, ...) for the pi coding agent via a self-contained local OpenAI-compatible gateway. Ported from Ebony-Vinyl/dsh-our-free-model.

umnaut Published 11 days ago MIT ~559 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

Reports 0

No reports yet.

Readme

pi-our-free-model

Free frontier models for the pi coding agent — no login, no sign-up, no API key. A self-contained local gateway provides models such as GLM-5.3, Kimi K3, DeepSeek V4.1 Flash, Nemotron 3.5 and many more, exposed to pi as an ordinary OpenAI-compatible provider.

This is a port. The original project is Ebony-Vinyl/dsh-our-free-model, a plugin for DeepSeek Harness (DSH). All credit for the model channels, the catalog and the gateway belongs to its author. This package adapts the upstream standalone service to pi's plugin API. See NOTICE.md.

Highlights

Zero configuration — install, then pick a model. No DSH, no Cordis, no DSH profile, no DSH credentials, no account.
Independent of DeepSeek Harness — the gateway is a plain local process with its own data directory (~/.our-free-model). It works on a machine where DSH was never installed.
Tools and streaming — tool calling and SSE streaming are verified working, which is what a coding agent needs.
Live model discovery — the model list comes from the gateway, enriched with each model's real context window, output limit, reasoning and vision support.
Runs out-of-process — so pi -p "..." exits cleanly, and multiple pi sessions share one gateway.

Requirements

pi coding agent 1.0.2 or later
Node.js 22.19+ or 24+ (matching the pi runtime)
Windows, macOS, or Linux

Install

pi install npm:pi-our-free-model

Or from a local checkout:

pi install ./pi-our-free-model

Then list the models:

pi --list-models our-free-model

Pick one interactively with /model, or for a single run:

pi --model our-free-model/stealth/glyph-cluster "hello"

Usage

# Interactive, then choose a model with /model
pi

# One-shot with an explicit free model
pi --model our-free-model/inclusionai/ling-3.1-flash "explain this repo"

# Status: gateway URL, data dir, and the current model list
/free-models

Model ids are used exactly as the gateway reports them, e.g. stealth/glyph-cluster, mimo-v2.6-flash-free, openrouter/free.

Configuration

Variable	Purpose	Default
`PI_OUR_FREE_MODEL_HOME`	Data directory for the gateway (settings, catalog, stats, API key)	`~/.our-free-model`

The gateway binds 127.0.0.1 only. Its API key is generated locally on first run and stored in <data dir>/settings.json as forwardKey.

How it works

pi process
  └── extensions/index.js        (this package)
        ├── spawns / adopts ─────► our-free-model gateway  (detached, 127.0.0.1)
        │                            └── vendor/packages/standalone/
        ├── GET /v1/models ──────► live model list
        └── pi.registerProvider("our-free-model", { api: "openai-completions", ... })

Two compatibility flags are required by this gateway and are set automatically:

supportsDeveloperRole: false — it rejects OpenAI's developer role
maxTokensField: "max_tokens" — it expects max_tokens, not max_completion_tokens

Notes and limitations

Free channels are best-effort. They are operated by the upstream project and their availability, rate limits and model rosters change without notice. Some models may be temporarily overloaded or retired. Treat this as a supplement to a paid provider, not a replacement.
The gateway is shared and stays running after pi exits, so the next start is instant. To stop it, end the node process whose pid is in <data dir>/service.lock, or remove that file after stopping the process.
A stale lock is handled. If a previous gateway was killed, the extension detects the dead owner, clears the lock and retries once.
This port is not affiliated with or endorsed by the upstream author.