Farkad AI — AI skill for Claude Code

farkad-ai

[](https://github.com/brmel/farkad-ai/actions/workflows/ci.yml) [](LICENSE)

The extraction engine behind [Farkad](https://farkad.web.app): it turns a sentence, a voice note or a photo into typed health entries. One routing call reads the capture and names the pillars it mentions (food, water, sleep…); one extraction call per tracked pillar it names then runs in parallel and answers in the schema the consumer supplies. Storage, auth and billing stay with the consumer.

Install

pip install "farkad-ai[google] @ git+https://github.com/brmel/farkad-ai"

The core imports no provider SDK. Pick one as an extra: `google` (Gemini on Vertex AI), `anthropic` or `openai`.

Usage

from google import genai

from farkad_ai import CaptureEngine, CaptureRequest, Logged, NothingToLog
from farkad_ai.models.vertex import VertexModel

engine = CaptureEngine(VertexModel(genai.Client(vertexai=True)), catalogue)
outcome = await engine.capture(CaptureRequest(tracked=(food, water), text="two eggs and a glass"))

match outcome:
    case Logged(extracted=extracted, refused=refused):
        print(extracted, refused)
    case NothingToLog(reason=reason):
        print(reason)

`catalogue` is every pillar routing may name (`PillarRegistryProtocol`). Each tracked pillar is a `TrackedPillar`: title, guidance, worked examples, entry schema and a `review` of each entry.

From the command line:

export GOOGLE_CLOUD_PROJECT=your-project GOOGLE_CLOUD_LOCATION=global
farkad-ai route --provider vertex --text "two eggs, a big glass of water, and I slept badly"

AI techniques

	Technique	Note
✅	Structured output	Every answer is validated against a Pydantic schema
✅	[Multimodal input](htt