Institutional learning layer for every agent in your company
HQ: San Francisco, CA, United States | Founded: 2026 | Employees: 1 | Stage: Seed (Y Combinator S26) | Website: https://www.tryglen.com/
Glen is a memory layer for AI agents. It gathers everything an organisation's agents have already done — sessions from Claude Code, Codex and Cursor, plus Slack messages, code reviews, tickets, documents and call transcripts — and when a new agent starts a task, it finds the relevant history and feeds it in, so the agent does not repeat work or ask for context it should already have. The company reports this cuts token spending by about 29% and shortens tasks by about 21%, by adding the right context rather than more of it. Glen doubles as a searchable record of company work: highlight a line of code and ask why it is there, and Glen answers from the original agent session rather than guessing. It also turns repeated team habits into reusable skills, lets two people work on the same document with agents that learn from each other, and can hand a session from one AI coding tool to another mid-task. Composio, the founder's former employer, rolled Glen out company-wide starting from two users, and more than a dozen other San Francisco startups use it. Founded in 2026 in San Francisco by sole founder Nikos Dritsakos, previously head of special projects at Composio.
Glen's insight is that the expensive part of AI work is not the model, it is re-explaining context the company already knows. Every agent session produces reasoning worth keeping, and nearly all of it is discarded when the session ends, so the next agent starts blind and burns tokens rediscovering it. Glen collects it all — sessions from Claude Code, Codex and Cursor, plus Slack, code reviews, tickets, docs and calls — and feeds the relevant pieces into new sessions before the model starts work. The reported numbers are the interesting part: about 29% less token spending and 21% faster task completion. The risk is sitting between three fast-moving AI platforms that all build their own memory features.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki