Pensieve: The Missing Onboarding Layer for AI Agents
Every company has the same problem with AI agents: the agent is smart but clueless about your business. It does not know who your customers are, which engineer owns what service, or that the Q2 pricing change means support tickets about billing will spike. You end up copy-pasting context into every prompt like a human RAG system.
Pensieve (pensieve.uk) takes a different approach. Instead of stuffing documents into a vector database and hoping for the best, it builds a living knowledge graph of your organization β mapping people, projects, decisions, customers, and how they all relate. When an agent queries Pensieve, it reasons over the full picture rather than retrieving isolated chunks.
The practical result: an AI agent that operates like a fully onboarded employee. It connects what customers say on calls with what engineers discuss on Slack with what usage data actually shows. The system surfaces connections that no single person catches because no single person has visibility across all those sources simultaneously.
Free to use, bring your own inference (Anthropic, OpenAI, or Google). Built on Neo4j for graph relationships, Supabase for backend, PostHog for analytics. Founded by Euan Cox and James Heavey in the UK. Launched on Product Hunt March 29.
The agent memory space is getting crowded, but most solutions are just glorified vector stores. Pensieve is betting that the right abstraction is not embeddings but relationships β a knowledge graph that understands organizational structure, not just document similarity. If agents are going to replace junior analysts, they need to understand the org chart, not just the docs.
← Back to all articles
Pensieve (pensieve.uk) takes a different approach. Instead of stuffing documents into a vector database and hoping for the best, it builds a living knowledge graph of your organization β mapping people, projects, decisions, customers, and how they all relate. When an agent queries Pensieve, it reasons over the full picture rather than retrieving isolated chunks.
The practical result: an AI agent that operates like a fully onboarded employee. It connects what customers say on calls with what engineers discuss on Slack with what usage data actually shows. The system surfaces connections that no single person catches because no single person has visibility across all those sources simultaneously.
Free to use, bring your own inference (Anthropic, OpenAI, or Google). Built on Neo4j for graph relationships, Supabase for backend, PostHog for analytics. Founded by Euan Cox and James Heavey in the UK. Launched on Product Hunt March 29.
The agent memory space is getting crowded, but most solutions are just glorified vector stores. Pensieve is betting that the right abstraction is not embeddings but relationships β a knowledge graph that understands organizational structure, not just document similarity. If agents are going to replace junior analysts, they need to understand the org chart, not just the docs.
Comments