# Lians recovery quickstart Set up the available free recovery layer, give an AI coding tool one safe project fact, then confirm that the current fact survives a fresh chat. The local setup requires no Lians account or provider API key. This quickstart proves local recovery and correction. The full Lians Guard workflow, including automatic lifecycle checkpoints, stale workspace detection, typed completion evidence, and trusted CI intake, remains a developer preview. ## 1. Choose a client ### Codex Install [`uv`](https://docs.astral.sh/uv/getting-started/installation/), then run: ```bash codex mcp add lians --env LIANS_MCP_ENABLED_TOOLS=remember,recall,list_memories,correct_memory,forget_memory -- uvx --from "lians-sdk[mcp]" lians-mcp ``` Restart Codex. Run `codex mcp list` or type `/mcp` in the terminal UI to confirm that `lians` is connected. ### Claude Code Run these commands inside Claude Code: ```text /plugin marketplace add Lians-ai/Lians /plugin install lians@lians-plugins ``` Restart Claude Code after installation. ### Cursor Use the [one-click Cursor installer](../integrations/cursor), review the local MCP configuration, and approve it in Cursor. For another MCP client, follow the [generic MCP setup](install.md#existing-ai-client-use-mcp). ## 2. Save one safe project fact In a project chat, say: ```text Remember that this project uses Python 3.12 and pytest. ``` Approve the `remember` tool if your client asks. Do not use passwords, API keys, personal data, or other secrets as test values. ## 3. Recall it in a fresh chat or second tool Open a new chat in the same project and ask: ```text What Python version and test runner does this project use? ``` Approve the `recall` tool if prompted. The answer should mention Python 3.12 and pytest without requiring the previous transcript. For the clearest continuity proof, save the fact in one connected tool and ask the question in another, for example, remember it in Claude Code and recall it from a fresh Codex task. Both tools must resolve to the same project and local Lians store. ## 4. Correct the memory Say: ```text Correct that memory: this project now uses Python 3.13 and pytest. ``` Open another fresh chat and ask the same question. Current-state recall should use Python 3.13 rather than presenting Python 3.12 as current. ## 5. Inspect or delete saved memory Ask your AI tool to list the project's Lians memories. To remove the test fact, ask it to permanently forget that memory and confirm the deletion when prompted. Deletion is intentionally explicit. Confirm that the correct memory reference is selected before approving it. ## Where local data is stored The basic MCP setup stores local memory in `~/.lians/mcp.db`. Set `LIANS_LOCAL_DB` in the MCP server environment to choose another location. Lians does not ask for your Claude, Cursor, or Codex password or provider API key. The first use may download a local semantic model. ## Preview the Guard tools from source Contributors can install the preview package from a clone: ```bash python -m pip install -e ./packages/lians-easy lians mcp ``` The preview MCP server includes `start_task`, `checkpoint_task`, `task_status`, `continue_work`, `configure_verification`, `verify_work`, and `verification_status`. Positive completion evidence must declare one of `measured_local`, `measured_ci`, or `human_confirmed`. Agent summaries and touched files remain recovery context and do not satisfy completion criteria. Read [the Lians Guard contract](lians-guard.md) before evaluating the preview. ## Troubleshooting - Confirm that `uvx` runs in a terminal. - Restart the AI client after changing its MCP configuration. - Confirm that the `lians` or `lians-memory` server is connected in the client. - Allow extra time during the first run while the local model initializes. - If a write times out during initialization, retry it after the server is ready; the timed-out write is not queued. For client-specific details, see the [Codex](../integrations/codex), [Claude Code](../integrations/lians-plugin), and [Cursor](../integrations/cursor) guides.