[ { "id": 36906, "name": "Memorable", "slug": "memorable", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/044e23bebf5afa8daf2de443d127002bbc57895a.png", "website": "https://www.memorable.sh/", "all_locations": "San Francisco, CA, USA; Remote", "long_description": "", "one_liner": "reducing agent reasoning into a graph search problem", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1789543087, "tags": [ "Developer Tools", "Infrastructure" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2027", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada", "Remote", "Fully Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": [ { "question": "What is your long-term vision? If you truly succeed, what will be different about the world?", "answer": "Our long-term vision is that AI agents accumulate experience instead of resetting after every session.\r\n\r\nIf Memorable succeeds, every completed task will make every future agent in an organization more capable. An agent facing a new goal will compose previously verified procedures, reason only about what is genuinely unfamiliar, and improve those procedures through use. Work that once required thousands of independent model decisions will become reliable execution over an evolving graph of operational knowledge.\r\n\r\nThe immediate implication is that we change what companies own. Today they rent intelligence from model providers while the experience generated by their agents disappears into logs. In the future, companies will own a durable memory of how their work gets done.\\\r\n\\\r\nThe greater vision however is a world where agents can pursue goals across weeks, months, and years; Memorable will be the runtime layer of abstraction for agents to do less menial work and more time thinking, planning, and orchestrating. **Long horizon tasks will be trivial.** Drift will be reduced 10x. \\\r\n\\\r\nAnd if everything goes perfectly, this **shared network** of **raw procedural intelligence** will get us one step closer to AGI." } ], "url": "https://www.ycombinator.com/companies/memorable", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2027/memorable.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/memorable-sh/about/?viewAsMember=true", "x_url": "https://x.com/memorablesh", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/044e23bebf5afa8daf2de443d127002bbc57895a.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/57e2408110cd21effc657ee481fca3ba7ce6ce7a.png" ], "primary_partner": { "name": "Kulveer Taggar", "url": null }, "founders": [ { "user_id": 2164442, "full_name": "Nikhil Krishnaswamy", "title": "Founder/CEO", "founder_bio": "CS @ Stanford\nBuilding navigation for agents @ Memorable\nnikhil@memorable.sh", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/nikhil-krishnaswamy/", "x_url": "https://x.com/nikhilk8754", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/ebb2609ed670c13f4da2e10851ae8caceca35a18.jpg" }, { "user_id": 3250603, "full_name": "Advaiyt Sane", "title": "Founder/CTO", "founder_bio": "CS @ CMU '29\nCreating Memorable, turning agent reasoning into graph search. ", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/advaiyt-sane-7a960a215/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/945086f34ce3b411a9ca8661006f418a4bab2718.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 115489, "title": "Memorable: graph procedural memory for AI agents. Stop re-teaching your agents!", "tagline": "Memorable gives AI agents procedural memory so they stop re-deriving the same workflow every time — cutting tool calls 60% and latency 40%.", "body": "# **_| TL;DR:_** Memorable records the tool calls an agent made to run a task, reuses that workflow on similar tasks instead of re-deriving through reasoning from scratch. –60% tool calls, –40% latency.\n\n\n\n**Memorable turns agent reasoning into a graph search problem.**\n\n1. We **capture your agents' traces** and store them as a lightweight graph.\n2. When a similar prompt arrives, we **extrapolate a workflow with a graph search.**\n3. We **inject a short pointer** (50 to 100 tokens) into the prompt.\n\n![uploaded image](/media/?type=post&id=115489&key=user_uploads/2164442/8a549d1f-423e-4011-90b8-d8e0145a1323)\n\n# **So the agent runs deterministically instead of re-deriving a plan every. single. time.**\n\n### **_Forget tokenmaxxing._**\n\n**We're graph-first and anti-text so you SAVE time, tokens, and money.**\n\n* **–60% tool calls** and **–40% latency** on [**QM's**](https://github.com/yc-software/qm) benchmarks\n* **–20% turns** in [**Gbrain**](https://github.com/garrytan/gbrain) bench, ran [**GStack’s**](https://github.com/garrytan/gstack) /investigate skill with **–98% tokens**\n\n**Do your agents need procedural memory?**\n\n* Memorable is for anyone running agents that do the same work more than once.\n* We've built for **coding, support, ops, research, company brain, and browser/computer-use agents.**\n* If you find yourself repeatedly having to correct or guide your agents, they need procedural memory!\n\n### **Imagine a world where your agents don’t reason.**\n\n* Most agent memory layers store facts using text: preferences, docs, past chats, while Memorable stores procedures: the steps an agent takes to achieve an outcome.\n* We are able to encode this graphically using nodes and edges, making it incredibly agent-friendly and token-efficient, allowing most agents to skip the preliminary planning step altogether.\n\n**Try Memorable, your agent will onboard you!**\n\n```\nnpx memorable-cli\ncurl -fsSL https://memorable.sh/install.sh | sh\n```\n\n* Compatible with **Claude Code, Codex, Cursor, and Claude Cowork**\n* You can try it out live in [**QM**](https://github.com/yc-software/qm) and [**Gbrain**](https://github.com/garrytan/gbrain) (thanks [**@Joshua France**](https://bookface.ycombinator.com/user/211454) &\n\n [**@Garry Tan**](https://bookface.ycombinator.com/user/678)!)\n* Procedures live encrypted in your own cloud, or run entirely local.\n\n![uploaded image](/media/?type=post&id=115489&key=user_uploads/2164442/59c2f411-3d34-482a-ba18-a03e14ecc4e2)\n\n**Memorable Team**\n\n* [**Nikhil Krishnaswamy**](https://bookface.ycombinator.com/user/2164442) \\[Stanford CS\\]: created an iOS MCP interning at AGI Inc, research published in COLM and IEEE\n* [**Advaiyt Sane**](https://bookface.ycombinator.com/user/3250603) \\[Carnegie Mellon CS\\]: built dev tools at Nvidia that create 50k AV tests a month\n\n**We are looking for design partners!**\n\n* Would love to chat with teams **deploying agents at volume** (high tool-call or token spend, repeated workflows) willing to **pilot as design partners.**\n* Open to **custom harness and framework integrations** for your interesting use cases\n\nEmail [info@memorable.sh](mailto:info@memorable.sh) and we'll spend a day solving your memory problems with you, plus:\n\n* LIFETIME of Memorable Pro FREE\n* Reach out to us about anything agent memory-related, whether or not you use Memorable :)\n\nand grab a slot: [**cal.com/team/memorable/30**](http://cal.com/team/memorable/30)\n\n[website](https://memorable.sh) | [docs](https://memorable.sh/docs) | [try memorable](https://memorable.sh)\n\nLet's build something Memorable!", "url": "https://www.ycombinator.com/launches/U2j-memorable-graph-procedural-memory-for-ai-agents-stop-re-teaching-your-agents", "votes": 15, "created_at": "2026-09-17T19:00:00.406Z" } ] } ]