LeadRescue AI

Self-healing lead operations: ingests ad-platform leads, delivers them into a CRM exactly once, recovers from failures automatically, reconciles lost webhooks, and explains incidents with an LLM.

Quick start (Docker)

docker compose up --build

That's all. The first build takes a few minutes; later starts take seconds.

What	Where
Mastra Studio (workflows, traces, agent, schedules)	http://localhost:4111
Health	http://localhost:4111/health/ready
Simulators (fake Meta Lead Ads + CRM)	http://localhost:4200

The stack starts in order: init generates secrets once into a volume, simulators starts, seed provisions two demo tenants, then app runs the production Mastra build (API, Studio, durable job worker and the reconciliation schedule). Data persists in Docker volumes across restarts.

Optional .env: to enable AI incident analysis, create .env with an Ollama Cloud key (https://ollama.com/settings/keys):

OLLAMA_API_KEY=your-key

Without a key, everything else works and incidents get deterministic assessments instead. To use a local Ollama daemon instead: LLM_PROVIDER=ollama-local and OLLAMA_BASE_URL=http://host.docker.internal:11434.

Send a lead (uses the simulator admin token from .env, or read it from the container: docker compose exec app sh -c 'grep SIMULATOR_ADMIN_TOKEN /config/secrets.env'):

T=<SIMULATOR_ADMIN_TOKEN>
curl -X POST localhost:4200/sim-admin/meta/pages/104729384710001/leads \
  -H "authorization: Bearer $T" -H 'content-type: application/json' \
  -d '{"fields":{"full_name":"Test Lead","email":"test@example.com","phone_number":"5125550100"}}'

Reset everything: docker compose down -v (removes the data and secrets volumes).

Local development (without Docker)

pnpm install && pnpm setup:env
pnpm sim          # terminal 1: simulators on :4200
pnpm seed         # once
pnpm dev:mastra   # terminal 2: Mastra dev server + Studio on :4111

Tests

pnpm test         # full suite, no LLM or servers needed
pnpm verify       # typecheck + lint + tests
pnpm ollama:smoke # one live LLM call (needs OLLAMA_API_KEY)
pnpm eval:ai      # live Incident Analyst evaluation across failure scenarios

# Or without Node on the host, inside the image:
docker compose run --rm --no-deps app node_modules/.bin/vitest run

Implementation status and verification evidence: docs/PROGRESS.md.