omp packages

pi-incoai0.1.3

Inco AI (inco.ai) model provider extension for the Pi coding agent

serggiu Published 6 days ago MIT ~531 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.

Did it work for you?

Result

One report per person. A new report replaces your old one.

Reports 0

No reports yet.

Readme

pi-incoai

Inco AI (inco.ai) model provider extension for the Pi coding agent.

Inco serves open-weight models (Kimi K3, GLM 5.3, MiniMax M3, DeepSeek V4.1, …) over an OpenAI-compatible API at https://api.inco.ai/v1. This package registers the incoai provider, adds the public model catalog, and refreshes it from GET /v1/models so workspace/private models show up too.

Install

pi install npm:pi-incoai          # from npm
pi install git:github.com/serggiu/pi-incoai   # from git
pi install /path/to/pi-incoai     # from a local checkout

or try it for one run without installing:

pi -e /path/to/pi-incoai

Authenticate

  1. Create an API key at https://platform.inco.ai/keys (it looks like sk-inco-…).

  2. Either set the environment variable in the shell that starts Pi:

    export INCO_API_KEY=sk-inco-...

    or sign in interactively:

    /login

    Pick incoai → Sign in with an API key → paste the key. Pi verifies the key against GET /v1/models before saving it, then refreshes the model catalog. Credentials are stored in ~/.pi/agent/auth.json.

Use

/model            # search for "incoai"

pi --model incoai/kimi-k3 "…"
pi --model incoai/glm-5.3:fast "…"

Model IDs keep their :fast suffix; Pi resolves the full id before treating a trailing :level as a thinking level.

Models

The bundled baseline catalog makes the public models available immediately and offline. Refreshing from GET /v1/models overrides the bundled metadata for the same model ids and adds any models the API returns that are not bundled — newly released public models and workspace/private models. Pricing, context length, and reasoning support come from GET /v1/models.

Models that are not part of the bundled snapshot are listed first (new public releases and workspace/private models), followed by the bundled public catalog; each group is alphabetical.

Model Context Max output Image input
kimi-k3 1.0M 131K yes
kimi-k3:fast 1.0M 131K yes
deepseek-v4.1-flash 1.0M 384K yes
deepseek-v4.1-flash:fast 1.0M 384K yes
glm-5.3 1.0M 131K no
glm-5.3:fast 1.0M 131K no
glm-5.3-flash 1.0M 131K yes
glm-5.3-flash:fast 1.0M 131K yes
minimax-m3 1.0M 512K yes
minimax-m3:fast 1.0M 512K yes

Notes

  • Inco is prepaid: requests return 402 when the balance is empty.
  • Inco exposes both OpenAI Chat Completions and Anthropic Messages. This extension uses Chat Completions, the surface its catalog metadata (reasoning_effort, reasoning_content) is documented for.
  • Models that advertise capabilities.reasoning_effort receive reasoning_effort. On the others Pi still surfaces streamed reasoning_content but sends no effort.
  • --list-models and pi auth check do not load extensions, so they show the bundled baseline catalog only. Opening /model triggers the live refresh.

Development

extensions/index.ts   # the provider extension

Pi loads extensions from the extensions/ directory of a package. Edit the file and run /reload in an active session, or restart Pi. To exercise the catalog mapping without a key, GET https://api.inco.ai/v1/models is public.

License

MIT