Description

Unsiloed Cookbook

Recipes for wiring Unsiloed‘s document AI — parse, extract, classify, split — into the AI provider, agent framework, or runtime of your choice.

The cookbook stays agnostic by design: each recipe is scoped to a single provider or skill family, so adding a new model, SDK, or agent runtime is a self-contained drop-in.

Prerequisites

• Unsiloed API key — sign up at unsiloed.ai
• Python 3.13+ for the notebook-based recipes (skill recipes need only curl and jq)

Getting Started

git clone https://github.com/Unsiloed-AI/cookbook.git
cd cookbook
Create your .env file
cp .env.example .env
Edit .env and add your API keys
For Python recipes
pip install -e .

Recipes

By AI provider

Recipes built around a specific model, SDK, or agent framework.

| Provider | Recipe | Description |
|———-|——–|————-|
| Claude | Tool Use | Give Claude structured access to Unsiloed via Anthropic’s tool-use API |
| Claude | Due-Diligence Agent | Upload a deal PDF → Unsiloed parse + Claude (LangChain) analyst + E2B sandbox that verifies every figure → risk-flagged diligence memo you can chat with |
| OpenAI | Agentic CLI | LangChain ReAct agent in a terminal REPL — plain-English document processing via Unsiloed |
| n8n | Unsiloed node | Community node to parse and extract documents with Unsiloed inside n8n workflows |
| Snowflake | SQL UDF | Python UDF that extracts documents from a Snowflake stage entirely in SQL, returning fields with confidence scores and citations |

Portable agent skills

Runtime-agnostic SKILL.md files — YAML frontmatter plus bash instructions — that plug into any skill-capable agent: Claude Code, Claude.ai, Hermes, OpenClaw, custom runtimes.

| Skill | Description |
|——-|————-|
| Unsiloed | Routes document reads through Unsiloed with confidence-scored output. Covers parse, extract, classify, split |
| Unsiloed HITL Extract | Guided field extraction with human-in-the-loop review: the agent asks which fields to pull, flags low-confidence values, and boxes them on an annotated copy of the document |

Repository layout

cookbook/
├── claude/         # Claude-specific recipes (tool use, agents, etc.)
├── openai/         # OpenAI-specific recipes (CLI, agents, etc.)
├── skills/         # Portable SKILL.md files for skill-capable runtimes
└── /     # Add new provider folders as siblings (gemini/, langchain/, ...)

Contributing

We welcome contributions! Pick the directory that matches what you’re adding:

• Provider-specific recipe (uses a particular SDK or API) → //.

Example: openai/responses-api/, gemini/function-calling/, langchain/extraction/.

• Portable agent skill (a SKILL.md that works across runtimes) → skills//.
• General guide that isn’t tied to one provider → top-level directory with a clear name.

For every recipe:

1. Include a README.md explaining what it does, prerequisites, and how to run it.
2. Use .env and python-dotenv (or shell export for skill recipes) for API keys — never hardcode them.
3. Clear notebook outputs before committing.
4. Open a PR with a description of what the example demonstrates and add it to the relevant table above.

Links

• Unsiloed Documentation
• API Reference
• Claude Integration Guide
• Support

License

MIT