Getting Started
What is Litica
Litica is a shared memory layer for AI workflows. It gives agents like Claude one place to keep context that sticks around between sessions. Agents can work from what they have learned, not just what fits in the current prompt.
Why it matters
Without memory, every agent session starts from nothing. Work that spans multiple steps or multiple agents loses context at every handoff. Litica gives each agent its own memory that lasts across sessions and builds up over time.
When you save a memory, Litica doesn't just store the words. It breaks the memory down into what it's about: the goal, the people and things involved, the mood, and the timing. Later, when an agent asks for context, Litica pulls up the related memories and ranks them by what fits the task and what has proven useful before.
Two ways in
Litica is hosted, so you do not need to run anything yourself. There are two paths to it — the same API key and the same memory behind both. Pick the one that fits how you work:
Connect an MCP client
Add one config block to Claude Desktop, Claude Code, or Cursor. Your agent gets memory tools automatically and uses them on its own.
Start with MCPPythonInstall the SDK
pip install litica and read, write, and share memory from code — every API route, one method each.
What your agent can do
On the MCP path, your agent has four tools available at all times:
- add_memory: save context, broken down into what it's about
- search_memory: pull back what fits the task right now, with the most useful memories first
- add_memories_batch: save a whole list of memories in one call
- add_documents: save whole documents; Litica splits them into memories for you
The agent calls these on its own when they help. You do not need to prompt it. Each agent has its own memory, identified by agent_id.
On the SDK path, the same memory sits behind litica.Client — one method per API route, including everything above plus namespaces, tracing, and the audit feed. The SDK reference covers the full surface.