Your AI will finally remember everything.
And act, not just answer.

Loomknot is a platform where AI agents get persistent memory, tasks, and access to your projects. Connect Claude, ChatGPT, Gemini, a local model, or any other agent — and turn a disposable chat into an autonomous assistant that works for you for weeks.

Works with any AI:ClaudeChatGPTGeminiOllama
Try for free

No card. 30 seconds to sign in. Connect your agent in 2 minutes.

Every AI today is a goldfish.

You explain context. List constraints. Bring them up to speed. Close the chat — and everything is forgotten.

ChatGPT added Memory. Claude made Projects. But that's their memory, not yours. Switch AI — start from zero. Work with a team — every agent is in its own bubble.

Now imagine: you connect a new agent — and it already knows everything. Your preferences, project history, tasks in progress, team decisions. Because the memory is stored with you, not the provider.

Three things that change everything

🧠

Memory that outlives any agent

Create a project. Add information — or let the agent do it. Tomorrow Claude 5 comes out or a new AI appears — connect it, and all memory is intact. The platform is your brain. The agent is a replaceable interface.

Three privacy levels: private (only you), project (team), public (via link). Protection at the database level — an agent physically cannot read someone else's private entries.

An agent that acts, not just answers

Your agent creates a task "find flights to Barcelona under €200." Searches for options. Records results. A week later reminds you: "check visa requirements." Tracks deadlines. Comes back with a report.

This is not a chat window. It's an autonomous assistant with a full audit log: what it searched, what it found, what's left to do.

👥

Multiple people. Multiple agents. One project.

Four friends plan a trip. Each has their own AI. Agents see the shared project memory, know their owner's private preferences, and suggest compromises when preferences conflict.

Not "a chat for four" — but four AIs working with shared data and negotiating with each other.

Get started in 2 minutes

1
📁

Create a project

"Barcelona 2026", "My Health", "Apartment renovation" — anything. The project automatically gets a main page and is ready to go.

2
🔗

Connect an agent

Generate an API key in project settings. Give it to your Claude, ChatGPT, local model, or any other MCP-compatible agent. The agent gets access to 24 tools.

3
💬

Work — the agent remembers

Everything you discuss and decide, the agent saves to structured project memory. Preferences, decisions, constraints, findings — all there for the next conversation.

4
🚀

The agent acts autonomously

Creates tasks, searches for options, compares prices, reminds about deadlines. You check results when convenient.

One product — dozens of applications

✈️ Travel

We were four planning Barcelona. Everyone connected their agent. My Claude knew I don't eat gluten and want a hotel with a pool. My friend's ChatGPT knew she needs a sea view. The agents found three hotels that work for everyone and created a vote. We just picked.
  • -Remembers each participant's preferences (privately)
  • -Creates tasks: search for flights, hotels, restaurants
  • -Finds options and records results to project memory
  • -Suggests compromises when preferences conflict
  • -Generates a final page with the trip plan

🏥 Health

"My Health" project is my medical assistant. The agent remembers all tests, medications, allergies. After two weeks it reminds me to buy medications. When I switched from ChatGPT to Claude — zero loss. The new agent connected and was immediately in context.
  • -Stores medical history in structured memory
  • -Creates reminders: "take tests in a month", "doctor appointment on the 15th"
  • -Private data is visible only to your agent — even if you add a spouse to the project
  • -When switching AI — all history and tasks remain

📚 Learning

I'm learning English. The agent remembers my level, weak spots, covered topics. Every week it creates a task with a new set of exercises. After a month I see progress — not because the agent is smart, but because it remembers.

🔨 Renovation

"Apartment Renovation" project. The agent knows the budget, timelines, contractor contacts. I say — "find a tiler for next week under $50 per sq ft." It creates a task, searches, comes back with options.

No lock-in. Your agent — your choice.

Loomknot works with any AI through the open MCP (Model Context Protocol) standard. Supported by Claude, ChatGPT, Gemini, local models (Ollama, LM Studio), and dozens of other agents.

Connect whichever you like. Tomorrow a new AI comes out — connect it instead. All memory, all tasks, all history — intact.

🟠

Claude

Native MCP connection. Perfect integration.

🟢

ChatGPT / Gemini

Via MCP adapter. Full access to memory and tasks.

💻

Local models

Ollama, LM Studio, llama.cpp — your data never leaves your computer.

🔧

Your agent

Any MCP-compatible bot. API key + 24 tools.

24 MCP tools. The agent doesn't chat — it works.

When you connect an agent to Loomknot, it doesn't just get "chat access." It gets a typed API to work with your data.

🧠

Memory

Read, write, search, delete. Three privacy levels. Semantic search by meaning, not keywords.

📄

Pages

Create, edit, generate. Each page is simultaneously UI for humans and data for the agent.

Tasks

Create, update, track. Priorities, deadlines, execution logs. The agent manages its own task manager.

📁

Projects

Info, members, settings. The agent understands the full project context.

🤝

Negotiations

Propose options, vote, find compromises. A mechanism for resolving conflicts between participants.

Your data — only yours. This isn't marketing, it's architecture.

Three privacy levels — not toggles in settings, but enforcement at the SQL query level.

Private memoryAnother user's agent physically cannot read it — the DB query filters by userId
API keysSHA-256 hashing, instant revocation, bound to a specific project
Project dataAccess only for members, 5 roles with different permissions
Public pagesYou decide what to show the world

We don't train AI on your data. We don't sell data. Data is stored on a dedicated server in Europe.

This is not a prototype. It's a working product.

Authentication (email + magic link)Done
API: ~50 endpointsDone
MCP: 24 agent toolsDone
Memory: 3 levels + cross-searchDone
Tasks: creation, tracking, logsDone
API keys (SHA-256, revocation)Done
Realtime updates (Socket.io)Done
Web interfaceActive development
Conflict auto-detectionPlanned
Semantic search (pgvector)Planned

Backend is 100% ready. We're polishing the interface and adding scenarios for early users.

Loomknot is for those who already get it

🚀

AI enthusiast

You already use Claude, ChatGPT, or a local model every day. It annoys you that every new chat starts from scratch. You want AI to remember context for weeks and do things while you sleep.

👥

Team with projects

You plan trips, events, renovations with friends or family. Everyone uses their own AI, and you want agents to work together, not each in their own corner.

🛠

Developer / hacker

You build your own agents and need infrastructure: persistent memory, task management, RBAC, API. You don't want to build from scratch — you want to connect via MCP and work.

Under the hood

🔗

MCP Protocol

Open standard by Anthropic. Not a proprietary API — the agent connects through a universal protocol supported by all major AIs.

🐘

PostgreSQL + pgvector

Not MongoDB, not Firebase. A relational database with vector search. Memory is structured, typed, and protected at the SQL level.

Realtime

Socket.io — instant updates. When the agent writes search results, you see them in the interface within a second.

🚀

Zero-downtime Deploy

Docker Swarm, rolling updates, automatic rollback. The platform doesn't go down during updates.

Stop explaining the same thing to AI every day.

Create a project. Connect an agent. Let it remember — and start acting.

Start for free

Email to sign in. No password. No card. First project — in 30 seconds.

FAQ

Yes, the platform is free right now. We're in early access and looking for first users to help shape the product.

Any agent supporting MCP (Model Context Protocol) — Claude, ChatGPT, Gemini, local models (Ollama, LM Studio), and any custom bot. The list grows every month.

Privacy is in the architecture, not the policy. Three access levels enforced at the SQL level. Data on a dedicated server in Europe. We don't train AI on your data.

Your data is yours. Full JSON export at any time. Memory is structured, not locked in a proprietary format.

Yes. The web interface lets you manage projects, memory, and tasks without a single line of code. Connecting an agent means copying an API key into your Claude, ChatGPT, or any other agent's settings.

Memory in ChatGPT is the provider's memory. It's tied to one AI, unstructured, and not controlled by you. In Loomknot, memory belongs to you, works with any AI, has three privacy levels, and a full API for automation.

LoomKnot — Your AI Will Finally Remember Everything