LaunchPilot

LaunchPilot is a hackathon-focused MVP for supervised, multi-agent product launches.

This repo is intentionally streamlined for a 36-hour delivery window:

• Next.js frontend (apps/web)
• FastAPI backend (apps/api)
• Postgres for app state
• Synchronous execution flow (no worker queue)
• Backboard-backed agent reasoning + memory threads

Core Product Slice

The shipped slice is:

1. Create project
2. Run research
3. Run positioning
4. Generate execution plan + assets
5. Prepare outreach batch
6. Approve send
7. Send emails (real Resend if configured, mock otherwise)

Repo layout

• apps/web: Next.js App Router UI
• apps/api: FastAPI API, SQLAlchemy models, Alembic migrations
• infra/docker: local Docker Compose for DB + migrate + API
• infra/scripts: utility scripts

Prerequisites

• Node.js 20+
• npm 10+
• Python 3.12+
• Postgres 15+ (or Docker)

Environment setup

1. Copy .env.example to .env
2. Choose auth mode:
   • AUTH_MODE=dev (recommended for hackathon speed)
   • AUTH_MODE=auth0 (strict JWT validation)
3. Configure Backboard:
   • Set BACKBOARD_API_KEY for the full multi-step agent pipeline.
   • API startup fails if BACKBOARD_API_KEY is missing.

Local development

1. Install web dependencies

npm run install:web

2. Install API dependencies

cd apps/api
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt

3. Run API migrations

From apps/api:

alembic upgrade head

4. Start API

From repo root:

uvicorn app.main:app --reload --app-dir apps/api

5. Start web

From repo root:

npm run dev

Docker quickstart

docker compose -f infra/docker/docker-compose.yml up --build

This starts:

• Postgres
• Migration job
• API

Note: frontend is not run by Docker Compose in this repo.

Auth modes

Dev mode (fastest)

Set:

AUTH_MODE=dev

Backend uses a local fallback user/scopes. No Auth0 API token verification is required.

Auth0 mode

Set:

AUTH_MODE=auth0
AUTH0_ISSUER=https://YOUR_TENANT.us.auth0.com/
AUTH0_AUDIENCE=YOUR_API_IDENTIFIER

Backend fails fast on startup if these are missing.

Backboard setup

Set:

BACKBOARD_API_KEY=your_backboard_api_key

Optional:

BACKBOARD_BASE_URL=https://app.backboard.io/api
BACKBOARD_LLM_PROVIDER=openai
BACKBOARD_MODEL_NAME=gpt-4o
BACKBOARD_MEMORY_MODE=On

Memory mode options are On, Auto, Readonly, Off. For this pipeline, On is the default to maximize persistent context retention across stage runs.

Backboard stage sessions are persisted per project and stage using project memory keys:

• backboard_research_assistant / backboard_research_thread
• backboard_positioning_assistant / backboard_positioning_thread
• backboard_execution_assistant / backboard_execution_thread

Each successful stage run also writes a structured stage snapshot into Backboard Memories (POST /assistants/{assistant_id}/memories) so future runs can retrieve richer context.

When a system prompt changes, the API automatically rotates the stored stage assistant/thread using a prompt fingerprint so new runs use the updated prompt.

Agent system prompts

All agent system prompts are stored in apps/api/app/prompts:

• apps/api/app/prompts/research_prompt.py (Research Agent system prompt)
• apps/api/app/prompts/positioning_prompt.py (Positioning Agent system prompt)
• apps/api/app/prompts/execution_prompt.py (Execution Agent system prompt)

These prompt constants are passed into Backboard assistant creation via:

• apps/api/app/agents/research_agent.py
• apps/api/app/agents/positioning_agent.py
• apps/api/app/agents/execution_agent.py

API overview

Base URL: http://localhost:8000/v1

Identity

• GET /me

Projects

• GET /projects
• POST /projects
• GET /projects/{project_id}
• POST /projects/{project_id}/brief
• POST /projects/{project_id}/sources
• GET /projects/{project_id}/memory
• GET /projects/{project_id}/activity

Research

• POST /projects/{project_id}/research/run (synchronous)
• POST /projects/{project_id}/research/advise (interactive follow-up on same stage thread)
• GET /projects/{project_id}/research

Positioning

• POST /projects/{project_id}/positioning/run (synchronous)
• POST /projects/{project_id}/positioning/advise (interactive follow-up on same stage thread)
• GET /projects/{project_id}/positioning
• POST /projects/{project_id}/positioning/select/{version_id}

Execution

• POST /projects/{project_id}/execution/plan (synchronous)
• POST /projects/{project_id}/execution/plan/advise (interactive follow-up)
• POST /projects/{project_id}/execution/assets (synchronous)
• POST /projects/{project_id}/execution/assets/advise (interactive follow-up)
• POST /projects/{project_id}/execution/contacts
• POST /projects/{project_id}/execution/email-batch/prepare (synchronous)
• POST /projects/{project_id}/execution/email-batch/prepare/advise (interactive follow-up)
• POST /projects/{project_id}/execution/email-batch/{batch_id}/send (requires approval)
• GET /projects/{project_id}/execution/state

Approvals

• GET /projects/{project_id}/approvals
• POST /approvals/{approval_id}/approve
• POST /approvals/{approval_id}/reject

Frontend route map

• /
• /login
• /app
• /app/projects
• /app/projects/new
• /app/projects/[projectSlug]
• /app/projects/[projectSlug]/research
• /app/projects/[projectSlug]/positioning
• /app/projects/[projectSlug]/execution
• /app/projects/[projectSlug]/approvals
• /app/projects/[projectSlug]/memory
• /app/projects/[projectSlug]/settings
• /app/settings/security

Legacy workspace routes still exist as redirects for compatibility.