Euan Cox

Our backend is a Python/FastAPI service with multiple worker processes (ingestion, agents, connectors, schedulers) and Railway handles all of it without us thinking about infrastructure. We deploy from a Dockerfile, Railway handles the rest. For a two-person team, the alternative was spending days on AWS/GCP config that doesn't move the product forward. Railway let us stay focused on building.

We have secrets spread across a Next.js frontend, a Python backend, multiple worker processes, and dev/staging/production environments. Doppler keeps all of that in sync without .env files scattered everywhere. For a product that connects to a dozen different APIs (Slack, Notion, GitHub, CRM, etc.), secret management is not optional.

We use PostHog across both our frontend and backend to understand how people interact with Pensieve's insight feed, workspace, and agent conversations. What we needed was analytics that a two-person team could self-serve without configuring dashboards for a week. Posthog is a pleasure to build with!

Supabase handles our auth, real-time subscriptions, and relational data alongside Neo4j for the graph layer. For a two-person founding team shipping fast, having auth, Postgres, real-time, and storage in one place meant we could focus on the hard problem (making businesses legible to AI) instead of stitching together infrastructure. The real-time subscriptions power our live updates when new insights surface. We went from zero to production without ever feeling like Supabase was the bottleneck.

Pensieve maps entire organisations into a knowledge graph so AI agents can reason over the relationships between people, projects, decisions, and customers. We needed a database that treats relationships as first-class citizens, and Neo4j is perfect for this.

Claude Code has become our go-to tool across the board. On the engineering side, its ability to build up a deep understanding of the repository is unparalleled. But I also use it constantly for non-technical work: strategy docs, launch planning, research. Pensieve is trying to do something similar for businesses: give AI full context so it can actually be useful, and we've taken a lot of learning from Claude Code.