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

# TypeScript Demos

> End-to-end Agent examples with OpenAI, Anthropic, and Convex storage

## Overview

The ICRL TypeScript package includes demo-ready examples that run with real API calls. These demonstrate the `Agent`, custom environments, and storage adapters. Run them from the `icrl-ts` directory.

## Prerequisites

From the monorepo root:

```bash theme={null}
cd icrl-ts
bun install
```

Required environment variables depend on the demo:

| Demo              | Required                                   | Optional                  |
| ----------------- | ------------------------------------------ | ------------------------- |
| OpenAI            | `OPENAI_API_KEY`                           | —                         |
| Anthropic         | `ANTHROPIC_API_KEY`                        | —                         |
| Support Triage    | `OPENAI_API_KEY`                           | —                         |
| Incident Response | `ANTHROPIC_API_KEY`                        | —                         |
| Web/Convex        | `OPENAI_API_KEY`, `NEXT_PUBLIC_CONVEX_URL` | `ICRL_CONVEX_DATABASE_ID` |

## Run All Core Demos

```bash theme={null}
cd icrl-ts
bun run examples:run
```

Runs: OpenAI, Anthropic, Support Triage, Incident Response.

## Run Individual Demos

```bash theme={null}
cd icrl-ts

bun run example:openai       # Basic Agent with OpenAIProvider
bun run example:anthropic    # Basic Agent with AnthropicProvider
bun run example:support      # Customer support triage automation
bun run example:incident     # Incident response playbook
bun run example:web         # Convex storage integration
```

## Run All (Including Web)

```bash theme={null}
cd icrl-ts
bun run examples:run:all
```

Includes the web-convex demo, which requires `NEXT_PUBLIC_CONVEX_URL` from a running Convex backend (e.g., `web-example`).

## Demo Index

| File                                 | Use Case                                                                         |
| ------------------------------------ | -------------------------------------------------------------------------------- |
| `examples/openai-demo.ts`            | Basic end-to-end `Agent` training with `OpenAIProvider` + `OpenAIEmbedder`       |
| `examples/anthropic-demo.ts`         | Basic end-to-end `Agent` training with `AnthropicProvider`                       |
| `examples/support-triage-demo.ts`    | Customer support triage: routing + structured reply (billing/shipping/technical) |
| `examples/incident-response-demo.ts` | On-call incident first-response playbook (severity, owner, first\_action)        |
| `examples/web-convex-demo.ts`        | Uses `web-example`'s `ConvexAdapter` for persistent storage                      |

## OpenAI Demo

Minimal Agent run with OpenAI:

```bash theme={null}
bun run example:openai
```

Uses `MathEnvironment` (trivial add-two-numbers task), `OpenAIProvider`, `OpenAIEmbedder`, and `FileSystemAdapter` for storage.

## Anthropic Demo

Same structure with Claude:

```bash theme={null}
bun run example:anthropic
```

Uses `AnthropicProvider` and `FileSystemAdapter`.

## Support Triage Demo

Customer support ticket routing and reply generation:

* **Environment**: `SupportTicketEnvironment` — routes to billing/shipping/technical based on content
* **Success**: Requires `priority`, `team`, and `reply` (≥ 40 chars) in the action
* **Use case**: Automating support triage with learned routing patterns

## Incident Response Demo

On-call incident first-response playbook:

* **Environment**: `IncidentEnvironment` — assigns owner (payments/backend/data) based on incident type
* **Success**: Requires `severity`, `owner`, and `first_action` in the action
* **Use case**: Standardizing incident response with learned playbooks

## Web/Convex Demo

Integrates with the `web-example` Convex backend:

* **Storage**: `ConvexAdapter` — trajectories live in Convex, same as the Next.js web demo
* **Setup**: Run `web-example` and set `NEXT_PUBLIC_CONVEX_URL`, or use a standalone Convex deployment
* **Optional**: `ICRL_CONVEX_DATABASE_ID` — if omitted, a demo database is created automatically