# What is Remiq

Remiq is a memory system for AI apps and agents. Your app saves what people and agents say. Before each model turn, Remiq gives your model the exact lines it should read, each with its date, its source and an id it can cite (`[M41]`). Your model writes the answer.

**Prism** is our model: it does the work inside Remiq, keeping, linking and finding what matters. You will see its name in the code (`prism_client.py`, `PRISM_API_KEY`, the `prism_…` MCP tools). Your model's key never reaches Remiq.

## What you get

- **The original words.** A save is kept as written, with who said it, when and where it came from. Remiq never rewrites it into a summary.
- **Changes, tracked.** When a value changes (a launch date, where someone works, a ticket's status), Remiq knows the current one and keeps the ones it replaced.
- **Privacy per person.** Each record can list its readers. A person sees only what they may see, and a hidden record is never hinted at.
- **Context for every turn.** One call returns `instructions` for your system prompt and `context` for the user turn, sized to your token budget.
- **Safe retries.** Send an `Idempotency-Key` and a retried write returns the first answer. You are never charged twice.

## Choose how you use it

| You want to | Start here |
|---|---|
| Add memory to your own app or agent | [Quickstart](/docs/quickstart/) |
| Give Claude, ChatGPT or another AI app one shared memory | [Connect your AI](/docs/tools/) |
| Give a coding agent the memory of past fixes | [Claude Code](/docs/claude-code/) and [Agent runs](/docs/agent-runs/) |
| Plug into LangChain, LlamaIndex or an agent SDK | [SDKs and frameworks](/docs/frameworks/) |

## What Remiq is not

- **Not a chatbot.** It returns context and evidence. Your AI, with your key, writes the answer.
- **Not a summarizer.** What your model writes can be saved as a note (`POST /v1/notes`), but a note is never used as evidence.