Stop explaining your work to AI every morning.

A new session should not erase yesterday's decisions. LiquidLM is persistent AI agent memory and RAG as a service.

The same loop runs whether the source comes from you or from an agent.

Everything becomes structure.

Files, notes, links, and recordings go into a vault. LiquidLM extracts the text and transcripts, then the entities, relationships, and timestamps that make them usable — not just a pile of text.

The vault gets sharper as it grows.

New knowledge links to what's already there. Temporal metadata and supersedes relationships quietly retire stale facts, and your own entity types teach the vault the vocabulary of your field. Every source you add improves the next answer.

RAG as a service — plus the agent memory sessions need.

Ask in the app with cited, grounded chat. Use LiquidLM as an MCP memory server for Claude Code or any MCP agent. Call the knowledge API from code, or search from the terminal. Four doors, one vault — shared agent memory, nothing to sync.

Estimate the time your team gets back.

Once a source is indexed, every lookup is faster than digging through files, chats, and someone's memory. This weighs that saved time against the cost of the plan your monthly ingestion calls for. Drag any slider to match your team — it's an estimate, not a guarantee.

The default of 5 minutes per lookup, at 6 lookups a day, is 30 minutes saved per person per day — about 6% of an eight-hour day, against the roughly 19–20% of the workweek the McKinsey Global Institute found knowledge workers spend searching for and gathering information (The social economy, MGI 2012). Nothing else is counted: no setup or curation time, and no credit for work you never had to redo.

Common questions.

What is LiquidLM?

LiquidLM is AI agent memory and RAG as a service. Drop in files, notes, links, and recordings; LiquidLM indexes them into a searchable knowledge vault your team uses in the app and your agents use through MCP, API, and CLI.

What problem does LiquidLM solve?

Knowledge lives scattered across files, chat scrollback, and someone's memory. Agents rediscover the same facts every run, and teammates dig for answers that were already found once. The bottleneck of AI-assisted work is no longer producing output — it's getting the right context to the right place.

What sources can I add?

Video, audio, images, PDFs, Office documents, plain text, notes, and links. Multimodal sources are transcribed or read and become searchable alongside everything else.

How do pricing and quotas work?

Three plans — Free, Pro ($20/mo), and Ultra ($200/mo) — sized mainly by how much you ingest per month. The quota terms:

Who owns my data?

You do. Knowledge is vault-scoped with per-vault credentials and an audit trail, forget is reversible, and your knowledge is exportable, so ownership stays clear. Subscriptions and metered usage are billed through Stripe.

How do I access my knowledge?

From the web app, from any MCP-compatible agent, from direct /v1 API calls, or from the terminal via the CLI — all reading the same indexed vault.

How is this different from Gemini Notebook (formerly NotebookLM)?

LiquidLM pairs a Gemini Notebook-style consumer experience with the developer surfaces that enterprise tools usually charge for — MCP, a REST API, and a CLI — at individual-friendly pricing, plus custom entity ontologies and temporal / stale-source guards.

How is this different from memory APIs like supermemory?

Developer-first memory layers ship SDKs and connectors but no consumer-grade place where your knowledge actually lives. LiquidLM pairs the same developer surface — MCP, REST API, CLI — with a real home for it: vaults, grounded chat with citations, and a browsable graph.

Why not an enterprise search platform?

Enterprise search platforms are connector-rich but priced per seat, usually behind a sales call. LiquidLM starts free, scales with usage rather than headcount, and keeps the full developer surface open on every plan.

Do you integrate with Google Drive, Notion, or calendars?

GitHub repository sync is the first automatic connector: choose a repository, branch, and optional directory before import. Google Drive, Notion, and calendars are not available yet. You can still add sources through the app, the /v1 API, the CLI, or any MCP agent.

Can my coding agent use LiquidLM as memory?

Yes. Point any MCP-compatible agent — Claude Code, for example — at a vault with scoped credentials. It can search existing knowledge and write new knowledge that you, your team, and your other agents see immediately.

What is RAG as a service?

RAG as a service means LiquidLM hosts the retrieval layer for you: ingest sources into a vault, keep knowledge current, and return cited context to AI agents through MCP or API — without running your own RAG stack.

Docs · Add to your assistant · Blog · Support