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

# Home

> The open-source TypeScript framework for building AI workflows and agents. Designed for Claude Code — describe what you want, Claude builds it, with all the best practices already in place.

Output is the open-source TypeScript framework for building AI workflows and agents. Designed for Claude Code — describe what you want, Claude builds it, with all the best practices already in place. Prompts, evals, tracing, cost tracking, orchestration, credentials. One framework. No SaaS fragmentation.

## Build AI using AI

Output is the first framework designed for AI coding agents. The entire codebase is structured so Claude Code can scaffold, plan, generate, test, and iterate on your workflows.

**Each workflow is a folder.** `workflow.ts`, `steps.ts`, `types.ts`, `prompts/`, `evaluators.ts` — everything your agent needs to understand context lives together.

**Ships with Claude Code integration.** Native commands, sub-agents, and a CLAUDE.md that teaches Claude the framework's patterns. Describe what you want in plain English. Claude builds it.

```bash theme={null}
# Tell Claude what you want
"Create a workflow that researches a company and generates a sales brief"

# Claude scaffolds the workflow, steps, prompts, and evals
# You refine from there
```

Beginners ship professional code from day one — the conventions teach best practices. Experienced AI engineers stop rebuilding the same infrastructure.

## Everything included

No more stitching together SaaS subscriptions. Output handles the fundamentals in one place:

**Where do prompts live?** In your repo as `.prompt` files — version-controlled, reviewable, deployable with your code. Not scattered strings. Not locked in external dashboards.

**How do you test non-deterministic code?** LLM-as-judge evaluators with confidence scores. Programmatic quality gates, not vibes.

**How do you track costs?** Every LLM call traces token counts and model info. Query execution history anytime.

**How do you handle secrets?** AI apps need a lot of API keys. Sharing `.env` files is risky, and coding agents shouldn't see your secrets. Encrypted credentials, scoped per environment, managed through the CLI.

**How do you scale?** Temporal is already there under the hood. You don't think about it until you need it.

## The Code

Two functions to organize your code: `workflow()` for orchestration, `step()` for I/O.

```typescript theme={null}
// workflow.ts - orchestration only
import { workflow, z } from '@outputai/core';
import { scrapeWebsite, summarizeContent } from './steps.js';

export default workflow({
  name: 'research_company',
  inputSchema: z.object({
    companyUrl: z.string().url()
  }),
  outputSchema: z.object({
    summary: z.string()
  }),
  fn: async ({ companyUrl }) => {
    const content = await scrapeWebsite(companyUrl);
    const summary = await summarizeContent(content);
    return { summary };
  }
});
```

```typescript theme={null}
// steps.ts - where I/O happens
import { step, z } from '@outputai/core';
import { generateText } from '@outputai/llm';
import { httpClient } from '@outputai/http';

export const scrapeWebsite = step({
  name: 'scrapeWebsite',
  inputSchema: z.string().url(),
  outputSchema: z.string(),
  fn: async (url) => {
    const client = httpClient({ timeout: 30000 });
    const response = await client.get(url);
    return response.text();
  }
});

export const summarizeContent = step({
  name: 'summarizeContent',
  inputSchema: z.string(),
  outputSchema: z.string(),
  fn: async (content) => {
    return generateText({
      prompt: 'summarize@v1',
      variables: { content }
    });
  }
});
```

## What's Included

<CardGroup cols={2}>
  <Card title="Prompt Management" icon="file-code">
    **Prompt Management.** `.prompt` files with LiquidJS templating. Version-controlled, reviewable in PRs, deployed with your code.
  </Card>

  <Card title="Multi-Provider LLM" icon="brain">
    **Multi-Provider LLM.** Anthropic, OpenAI, Azure, Vertex AI, Bedrock. Switch providers by changing one line in your prompt file.
  </Card>

  <Card title="LLM-as-Judge Evals" icon="scale-balanced">
    **LLM-as-Judge Evals.** Built-in evaluators with confidence scores and reasoning. Test non-deterministic code programmatically.
  </Card>

  <Card title="Tracing and Cost Tracking" icon="chart-line">
    **Tracing and Cost Tracking.** Every operation traced automatically. Token counts, costs, latency. JSON on disk, zero config. You own your data.
  </Card>

  <Card title="Durable Orchestration" icon="rotate">
    **Durable Orchestration.** Temporal under the hood. Automatic retries, workflow history, replay on failure. You don't think about it until you need it.
  </Card>

  <Card title="Encrypted Credentials" icon="lock">
    **Encrypted Credentials.** AES-256-GCM encrypted secrets, scoped per environment and workflow. No more sharing `.env` files or external vault subscriptions.
  </Card>
</CardGroup>

<Card title="Get Started" icon="rocket" href="/start-here/getting-started">
  **Get Started.** Set up your first Output project in 5 minutes.
</Card>