Why Lens Prism

Lens Prism is an agent, not a chatbot. It has tools and it decides when to use them, and it keeps state across conversations.

A persistent identity

A Prism agent isn't a stateless chat session. Its behavior is defined by a small set of markdown documents — Soul, User, Memory, Vision, Agent Guide, Heartbeat, and Bootstrap — that stay in its system prompt and travel with it across restarts. You shape them through conversation, and the agent maintains the ones that capture what it learns. See Agent identity.

Memory that survives conversations

The agent remembers across conversations through two layers: a curated long-term document it always keeps in context, and an embedding-indexed store it searches by meaning when something becomes relevant. Both persist in the embedded database. See Memory.

Proactive, not only reactive

The agent runs a heartbeat monitoring loop on a schedule, puts one-off and recurring work on its own calendar with scheduled tasks, and reacts to external events pushed to webhooks. It decides on its own whether a finding is worth telling you about.

Real work over a shell

The agent acts on a real shell — running commands, reading and writing files, and delegating larger coding jobs to a Claude Code sub-agent. It extends itself with installable skills. See Tools.

One process, one dependency

Prism runs as a single self-contained process. One container, one embedded SQLite database, no message broker or external datastore to stand up. The database holds identity, conversation history, memory, scheduled tasks, and Slack mappings, so backing up one file backs up the whole agent. The only hard dependency is an LLM provider. See Installation.