Website · Install · Docs · Pricing · Issues · Star Lians · Follow @Lians-ai

Memory for any AI agent.

Lians gives AI agents durable memory across chats, sessions, tools, and models. Your agent can remember a useful fact now and recall it when it matters later.

• Works with the AI you already use through MCP, plugins, or an SDK.
• Runs locally by default with SQLite and no Lians account or API key.
• Keeps memory focused by returning a small, relevant set of current facts.
• Stays provider-neutral so memory is not trapped inside one model vendor.

Lians is a memory layer, not another assistant. Your agent and model stay the same; Lians gives them a place to remember.

Start free. Upgrade only when it saves you time.

• Community is free: run Lians locally or self-host it with no account, API key, or software license fee.
• Lians Personal is $10/month: get a managed account, a private hosted memory workspace, 100,000 writes and 50,000 recalls each month, export and deletion controls, and email setup support. Cancel anytime.

First-customer setup: the first Personal customer gets a 30-minute founder-led setup session at no extra cost. We will connect one supported AI tool and run the two-chat memory test together. Purchase only when managed memory solves a real problem for you.

Get Lians Personal for $10/month or install the free local version through MCP.

Pick the AI you already use

• Cursor: use the one-click MCP installer.
• Claude Code: paste the two plugin commands.
• Codex app, CLI, or IDE: run the one-command MCP setup.

All three routes use the same free local memory by default. After setup, try the three-minute, two-chat memory challenge.

Running a club, hackathon, class, or campus developer community? Use the student and community kit for a ready-made workshop, project track, judging rubric, and shareable announcement.

Lians does not change the underlying model or promise fewer tokens on every task. It can avoid resending old conversation when a small relevant recall is enough; verify the result on your own workflow.

▶ Watch the 33-second demo: remember, recall, and confirmed deletion

Simplest setup: Lians Personal

Choose Personal when you want Lians managed for you instead of maintaining a local memory service. It includes a private hosted workspace, 100,000 writes and 50,000 recalls each month, export and deletion controls, and email setup support for $10/month. Cancel anytime.

Start Lians Personal with founder-led setup

The standalone desktop installer is still a technical preview. Signed Windows and notarized macOS downloads are not in the current GitHub releases, so the project does not present a nonexistent or unsigned download as the normal-user path. Developers evaluating the preview can use the Lians Easy source guide.

Free local setup: add memory through MCP

Use this path when you prefer a package-managed MCP server or want the full temporal and governance engine.

Install uv, then add this server to your agent's MCP configuration:

{
  "mcpServers": {
    "lians": {
      "command": "uvx",
      "args": ["--from", "lians-sdk[mcp]", "lians-mcp"],
      "env": {
        "LIANS_MCP_ENABLED_TOOLS": "remember,recall,list_memories,correct_memory,forget_memory"
      }
    }
  }
}

Restart your agent and try two prompts in separate chats:

Remember that this project uses Python 3.12 and pytest.

What Python version and test runner does this project use?

On a clean machine, the first memory tool may download and initialize the local semantic model. Lians reports that warmup to MCP clients that support progress. If it is still running after 90 seconds, the tool returns a retryable error and does not queue the write; keep the MCP server running and retry shortly. Model files use the Hugging Face cache controlled by HF_HOME.

Local MCP memory is stored in ~/.lians/mcp.db. The starter configuration exposes the basic loop plus the controls needed to trust it:

Tool	What it does
remember	Store one durable fact, preference, constraint, or decision.
recall	Retrieve a small set of relevant, current memories.
list_memories	Inspect what Lians currently knows.
correct_memory	Replace a stale fact without hiding its history.
forget_memory	Permanently erase one memory after explicit confirmation.

Use the exact setup guide for Cursor, Gemini CLI, Claude Code, or OpenCode, or Codex. Remove LIANS_MCP_ENABLED_TOOLS when you want the advanced temporal and audit tools too.

How it works

you → your AI agent → remember / recall → Lians → local SQLite

1. You or your agent explicitly saves something worth keeping.
2. A later session asks Lians for memory related to the current task.
3. Lians returns bounded context instead of replaying every old conversation.
4. When a fact changes, Lians can supersede the stale version instead of sending both versions back to the model.

No model provider owns the memory. You can point another compatible agent at the same Lians store and continue from the same context.

Use Lians in Python

For an application, notebook, or agent loop that needs in-process memory:

pip install "lians-sdk[local]"

from datetime import datetime, timezone
from lians import LocalLiansClient

memory = LocalLiansClient(db_path=".lians/memory.db")

memory.add(
    agent_id="my-agent",
    content="The project uses Python 3.12 and pytest.",
    event_time=datetime.now(timezone.utc),
    metadata={"project": "demo", "topic": "tooling"},
)

result = memory.recall(
    agent_id="my-agent",
    query="Which Python version and test runner should I use?",
)

for item in result["memories"]:
    print(item["content"])

Local mode needs no server, Docker container, or API key. The first run may download the local embedding model.

Install from this repository

git clone https://github.com/Lians-ai/Lians.git
cd Lians
python -m pip install -e "agentmem/sdk/python[local,mcp]"

That editable install includes the local Python client and the lians-mcp entry point. See the full install guide for TypeScript, Go, Java, C, framework adapters, and self-hosting.

Choose the interface that fits

You want to...	Start with
Use managed private memory without running a server	Lians Personal — $10/month
Give an existing AI client free local memory	MCP setup
Add local memory inside Python	LocalLiansClient
Evaluate the desktop installer preview from source	Lians Easy
Connect Python or TypeScript to a Lians server	Language SDKs
Use Pydantic AI, LangChain, LangGraph, CrewAI, OpenAI Agents, or AutoGen	Framework integrations
Run the full service yourself	Self-host Lians

Cloned the repository and unsure which package or folder is current? Read Supported paths and repository status before choosing an SDK, plugin, preview, or legacy tree.

Why Lians

Most memory demos store text and run vector search. Lians also handles the problems that appear when an agent keeps memory for more than a few sessions:

• Current over stale: corrected facts can supersede earlier versions.
• Small over noisy: recall is bounded so the model gets useful context.
• Local over locked-in: local mode keeps data on your machine.
• Portable over provider-specific: MCP and SDKs work across agent stacks.
• Inspectable over opaque: memories can retain timestamps, sources, and lineage.

Advanced capabilities

Lians also supports point-in-time recall, conflict inspection, memory lineage, tamper-evident audit history, governed erasure, information barriers, and decision reconstruction. These capabilities are available when a project needs them; they are not required to get started.

• Memory engine
• Decision evidence
• Security model
• Benchmarks and reproducible evidence
• Community and managed product boundary

Repository map

agentmem/src/lians/          Core engine and HTTP service
agentmem/sdk/python/        Python SDK, local client, and MCP server
agentmem/sdk/typescript/    TypeScript SDK
packages/lians-easy/        Dependency-free desktop runtime and installer
integrations/               Agent and framework integrations
plugins/                    Installable agent plugins
docs/                       Setup, architecture, security, and operations

Development

git clone https://github.com/Lians-ai/Lians.git
cd Lians
python -m pip install -e ".[dev]"
python scripts/test_all.py

Focused test runs and development conventions are in CONTRIBUTING.md. Published package and registry versions are tracked in docs/published-release-status.json.

Community

• Ask a question or report a bug in GitHub Issues.
• Request a new agent or framework integration with the integration template.
• Read the security policy before reporting a vulnerability.

If Lians is useful to you, star the repository. It helps other agent developers find the project.

License

Apache 2.0 — see LICENSE.