Tasks SQLite Example

A streaming AI agent for project and task management, powered by Mimic synthetic data and the Vercel AI SDK. No Docker or external database required — SQLite is file-based.

The agent streams responses using the AI SDK UI Message Stream protocol, and a minimal Next.js chat UI is included to interact with it in the browser.

Prerequisites

• Node.js >= 22
• sqlite3 CLI (pre-installed on macOS; apt install sqlite3 on Debian/Ubuntu)
• An Anthropic API key

Quick Start

1. Install dependencies

# Install the Mimic CLI and SQLite driver
npm install

This installs @mimicai/cli and better-sqlite3 from the root package.json.

2. Create the database

./init-db.sh

This creates tasks.db from schema.sql with tables for projects, tasks, labels, task_labels, and comments.

3. Set your API key

Create a .env file in this directory:

ANTHROPIC_API_KEY=sk-ant-...

Then export it so the CLI can use it:

export $(cat .env | xargs)

4. Generate synthetic data

npx mimic run

This uses the two personas defined in mimic.json (a busy developer and a project manager) to generate realistic data blueprints via the Anthropic API. Blueprints are saved to .mimic/data/.

5. Seed the database

npx mimic seed --verbose

This inserts the generated data into tasks.db. You can inspect what was created:

npx mimic inspect schema   # show tables and columns
npx mimic inspect data     # show row counts per table

6. Start the agent

cd agent
npm install
npm start

The agent starts on http://localhost:3002. You can test it with curl:

curl -X POST http://localhost:3002/chat \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "What projects are active?"}]}'

7. Start the chat UI

Open a new terminal (keep the agent running):

cd ui
npm install
npm run dev

Open http://localhost:3000 in your browser. You should see a chat interface — type a message like "What projects are active?" and hit Send.

How It Works

Browser (localhost:3000)
  └─ useChat hook sends messages to /api/chat
       └─ Next.js proxy converts UIMessage → ModelMessage format
            └─ Forwards to agent (localhost:3002/chat)
                 └─ Agent calls Anthropic API with tools
                      └─ Tools query SQLite database
                           └─ Streams response back through the chain

The UI uses the AI SDK useChat hook from @ai-sdk/react. The Next.js API route at ui/src/app/api/chat/route.ts proxies requests to the agent, converting the UIMessage parts format to simple {role, content} messages that the agent expects.

Testing with curl

# Simple message
curl -X POST http://localhost:3002/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Show me all blocked tasks"}'

# Multi-turn conversation
curl -X POST http://localhost:3002/chat \
  -H "Content-Type: application/json" \
  -d '{"messages": [
    {"role": "user", "content": "List all projects"},
    {"role": "assistant", "content": "Here are your projects: ..."},
    {"role": "user", "content": "Show me tasks in the first project"}
  ]}'

# Health check
curl http://localhost:3002/health

Configuration

Agent environment variables

| Variable | Default | Description | |---|---|---| | PORT | 3002 | Agent HTTP server port | | MODEL | claude-haiku-4-5 | Anthropic model to use | | DB_PATH | ../tasks.db | Path to the SQLite database | | ANTHROPIC_API_KEY | (required) | Your Anthropic API key |

UI environment variables

Set these in ui/.env.local:

| Variable | Default | Description | |---|---|---| | AGENT_URL | http://localhost:3002 | Agent server URL |

Agent Tools

The agent exposes six tools to the LLM:

| Tool | Description | |---|---| | list_projects | List all projects with task counts, filter by status | | search_tasks | Keyword search across task titles and descriptions | | get_task_details | Full details for a single task including labels | | get_project_tasks | All tasks for a project, ordered by priority | | get_task_comments | Chronological comments on a task | | get_blocked_tasks | All blocked tasks, optionally filtered by project |

Schema

| Table | Description | |---|---| | projects | Top-level containers with name, description, and status | | tasks | Work items with status, priority, assignee, dates, and hour tracking | | labels | Reusable tags with color codes | | task_labels | Many-to-many join between tasks and labels | | comments | Chronological discussion threads on tasks |

Task statuses: todo → in_progress → review → done (or blocked at any stage)

Task priorities: low | medium | high | urgent

Cleanup

rm tasks.db                  # remove the database
npx mimic clean --yes        # also remove generated blueprints