pi-ainetcafe-provider
v0.1.0
Published
ainetcafe provider for pi: Kimi K3 served from ainetcafe's own cluster at native MXFP4 ($2.10 / $10.50 per M tokens), OpenAI-compatible, with thinking-level mapping and live model discovery.
Maintainers
Readme
pi-ainetcafe-provider
ainetcafe provider for pi. Kimi K3 served from ainetcafe's own cluster at the released MXFP4 precision, $2.10 / $10.50 per million tokens, $0.30 cached input, OpenAI-compatible.
Install
export AINETCAFE_API_KEY=sk-... # Token Management at https://microquickjs.com
pi install npm:pi-ainetcafe-provider
pi
/model # pick ainetcafe/Kimi-K3Or without installing anything: add the provider to ~/.pi/agent/models.json as described in the pi guide.
What it does
- Registers
ainetcafewith base URLhttps://microquickjs.com/v1(override withAINETCAFE_BASE_URL). - Declares
Kimi-K3with 256K context, text + image input, reasoning on, and pi's thinking levels mapped toreasoning_effortlow / high / max. K3 always thinks, so theofflevel is marked unsupported. - Queries
/v1/modelsat startup (5 s timeout) so other models on the account appear too, with conservative defaults.
Verified endpoint behaviour
Checked with curl before writing the compat block: reasoning_effort is honoured, reasoning comes back as reasoning_content, function tool calls and streaming with usage work, image input works, thinking: {type: "disabled"} is accepted but does not turn thinking off.
Live availability probe: https://ainetcafe.com/k3/status.html
License
MIT
