> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usehindsight.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Build Your Custom AI

> Turn Hindsight's tools into repeatable AI you and your team can run on demand

Once your deals are analyzed and your library is seeded, Hindsight's AI can reason over all of it. This page is about going from one-off questions to **repeatable, custom AI** — capabilities you build once and your whole team runs.

There are two ways to do that, and they're the same idea at two altitudes:

<CardGroup cols={2}>
  <Card title="Recipes — your in-app Skills" icon="wand-magic-sparkles" href="/guides/recipes">
    Reusable, parameterized prompts you save and schedule inside Hindsight. Think of a Recipe as a **Claude Skill that lives in the app** — package a task once, then run it on demand or on a cadence.
  </Card>

  <Card title="Skills in Claude — over MCP" icon="plug" href="/integrations/mcp">
    Connect Hindsight's MCP server to Claude and build a **Claude Skill** that calls Hindsight's tools directly. Same verified data, orchestrated by your own agent in whatever workflow you already use.
  </Card>
</CardGroup>

***

## How Hindsight's AI works

The hard part of useful AI is getting the right context into the model at the right time — *context engineering*. Hindsight handles this by exposing research tasks (like "find objections vs. Acme in lost deals this quarter") as **tools** the agent invokes from natural language.

Simple queries just work — Hindsight plans the response and picks the right tools. But knowing the tools lets you be precise and chain them into more complex tasks:

* **Deal research** — `select_deal`, `search_across_deals`, `search_deal_documents` to find and read deals.
* **Competitive intelligence** — search your assets, saved intel, and knowledge base; live web search and scrape.
* **Quantitative analysis** — build datasets and analyze the deal grid for win-rate and trend questions.

Behind the scenes, Hindsight routes each query to specialized sub-agents. For the full tool reference, prompting patterns, and worked examples, see the [AI Agent guide](/guides/prompting).

***

## Build a Recipe (your in-app Skill)

Recipes let you package a working prompt — with variables — and save it so anyone on your team can run it. **A Recipe is Hindsight's in-app version of a Claude Skill:** you define the task once, and it becomes a reusable capability.

1. Go to **Chat** in Hindsight and click **Browse Recipes**.
2. **Remix** an existing recipe or click **Create Recipe**.
3. Add **variables** (competitor name, deal filters, date range) so the recipe is reusable across runs.
4. Run it to test, then save and share.

Once a Recipe works, schedule it with **Workflows** to deliver to Slack or email, write back to your library, or push to other tools via Zapier. Full details — recipe types, variables, and best practices — are in the [Recipes guide](/guides/recipes).

<Tip>
  Keep each Recipe focused on one job and give variables clear names. A tight, single-purpose Recipe produces far better output than one that tries to do everything.
</Tip>

***

## Strong recommendation: build a Skill in Claude over MCP

Recipes are the fastest way to get repeatable AI inside Hindsight. But the most powerful path is to **build a Claude Skill that uses Hindsight's tools directly over [MCP](/integrations/mcp)**.

Because Hindsight exposes its verified deal layer as MCP tools, you don't have to rebuild any retrieval or analysis — you point Claude at the same `select_deal`, `search_across_deals`, `search_competitive_intel`, and dataset tools the in-app agent uses, and let Claude orchestrate them inside a Skill you author. That gives you:

* **Your own orchestration** — Claude decides how to chain Hindsight's tools for your task, with your instructions and guardrails.
* **Composability** — combine Hindsight tools with everything else Claude can reach (other MCP servers, files, code).
* **Portability** — the Skill runs wherever you use Claude, not just inside Hindsight.

If you're building custom AI on top of your GTM data, **start the Skill in Claude and give it Hindsight's MCP tools.** Generate a workspace-scoped key from **Settings → API Keys**, then follow the [MCP setup guide](/integrations/mcp).

***

## What's Next

<CardGroup cols={3}>
  <Card title="Connect over MCP" icon="plug" href="/integrations/mcp">
    Wire Hindsight into Claude and build a Skill on your verified deal layer.
  </Card>

  <Card title="Recipes Guide" icon="wand-magic-sparkles" href="/guides/recipes">
    Recipe types, variables, scheduling, and best practices.
  </Card>

  <Card title="AI Agent Tools" icon="robot" href="/guides/prompting">
    The full tool reference and prompting patterns for precise results.
  </Card>
</CardGroup>