Docs

Quickstart

Two ways in: an MCP connector for any MCP client, or a REST API for everything else. Both reach one memory graph.

Streamable HTTP
https://api.kijito.ai/mcp
REST base
https://api.kijito.ai/api

MCP is an open standard. Kijito works in any MCP-compatible client. The steps below are the worked examples for the popular ones.

  1. Open your client's connector or MCP-server settings.
  2. Add a server with the Streamable HTTP URL below - it works in every modern MCP client.
  3. Sign in with your Kijito account when the client redirects you.
  4. Your agent now has remember, recall, dream and the rest of the tools in every conversation.
server url
https://api.kijito.ai/mcp

Your first five minutes

Kijito is long-term memory for AI agents: a graph your assistant writes to and recalls from, so it continues across conversations instead of starting cold each time. Here is the whole loop, start to finish.

1. Create an account

Sign up with an email and password. Your memory content is encrypted under a key derived from that password, so keep it (and your recovery codes) safe - see Security & trust. New accounts start with a small memory allowance; verify your email to lift it to the full free-tier cap.

2. Connect your client

Add Kijito to your assistant using the connector above. In an MCP client (Claude, ChatGPT, Cursor, and others) you paste one Streamable HTTP URL and sign in; your agent then has remember, recall, dream and the rest of the tools in every conversation. Prefer to build directly? Use the REST tab for a token-and-curl setup.

3. Remember something

Just talk. When something matters - a decision, a fact, a preference, where you left off - your agent calls remember to store it. You can also nudge it: "remember that we chose aqua as the primary accent." One atomic idea per memory works best; the graph links related memories for you.

4. Recall it later

In a new conversation, ask about it - "what did we decide about color?" Your agent calls recall, which searches by meaning (not keywords), pulls in the memories connected to the best matches, and returns just the few that are relevant. That is the payoff: continuity without re-reading the whole history.

5. Watch it in the graph

Open your graph view to see your memories as nodes, the edges between them, and the themes that emerge. Over time a scheduled "dreaming" pass connects related ideas, lets unimportant ones fade, and keeps recall sharp - you do not manage any of it.

How your memory stays useful

You don't manage any of this. Your AI does it as you talk:

  • Saves what matters as you go: decisions, facts, preferences, where you left off.
  • Corrects itself when things change. The old version is kept for history, never deleted, so you can always see how its understanding evolved.
  • Tidies up over time. Unimportant memories fade and related ones get connected (the "dreaming" pass), so recall stays sharp.
  • Curates as a chat wraps up. A good agent reviews what it learned, fixes anything now out of date, and saves what the next conversation will need.

The core tools

Kijito exposes a full toolset over MCP, but day to day an agent reaches for a small core - and recall returns only what matters, not a wall of definitions.

startupSession bootstrap: identity, recent memories, and active goals in one call.rememberStore a memory. One call, persisted forever, recalled by meaning.recallFind the few memories that relate, by meaning, and the ones connected to them.correctSupersede a memory without losing the history. The old version is kept.dreamRun a consolidation pass: prune, strengthen, cluster, link.send / inboxMessage other agents on the same graph. Coordinate a hive.

Keep going: read Concepts to understand how memory, edges, themes, and dreaming fit together, or jump to the API reference. Questions in the meantime: [email protected].