Community integrations for Nowledge Mem

---

The canonical source of truth for all integrations is integrations.json. Capabilities, install commands, transport, tool naming, thread save methods, and the user-facing autonomy contract are tracked there. Update the registry first when adding or modifying integrations.

The autonomy contract uses one shared language across integrations:

• automatic: the host/plugin enforces it through hooks or lifecycle wiring
• guided: the package/rules/skills strongly teach it, but the model still decides
• manual: it only happens when the user or agent asks directly

This keeps one critical distinction honest for fresh users: having tools available is not the same thing as getting autonomous memory behavior.

For behavioral guidance (when to read Context Bundle, use Working Memory fallback, search, save, and route ambient spaces), see shared/behavioral-guidance.md. For plugin authoring rules, see docs/PLUGIN_DEVELOPMENT_GUIDE.md.

For end-user customization that survives updates, see docs/USER_OVERRIDE_GUIDE.md. The short version: do not edit installed plugin files; use the host's own instruction files when that host supports them.

For multi-agent orchestrators that launch Codex, Claude Code, OpenCode, or other child CLIs, set NMEM_AGENT_ID=<agent-slug> in each child process. Add NMEM_SPACE=<space> only when that run should override the AI Identity's default space. NMEM_HOST_AGENT_ID is for advanced external aliases, not a second required identity variable. The child plugin still reports its real runtime as source_app; the env var selects the right Nowledge AI Identity through Context Bundle.

Each directory is a standalone integration. Pick the one that matches your tool.

Add to your tool's MCP settings:

{
  "mcpServers": {
    "nowledge-mem": {
      "url": "http://127.0.0.1:14242/mcp/",
      "type": "streamableHttp"
    }
  }
}

See mcp.json for the reference config.

For remote Mem, configure this machine once with nmem config client set url ... and nmem config client set api-key ..., then generate the exact host config:

nmem config mcp show --host cursor
nmem config mcp show --host codex
nmem config mcp show --host gemini-cli
nmem config mcp show --host craft-agent
nmem config mcp show --host codebuddy

Direct MCP clients do not read ~/.nowledge-mem/config.json automatically; paste the generated block into the host's own MCP settings.

• Nowledge Mem running locally
• nmem CLI on your PATH: if Mem is running on the same machine, install it from Settings > Preferences > Developer Tools > Install CLI in the app, use pip install nmem-cli for a standalone setup, or on Arch Linux install the nmem-cli AUR package with yay -S nmem-cli or paru -S nmem-cli

nmem status   # verify Nowledge Mem is running

Spaces are optional. Most integrations can stay on Default and never mention them.

If a host already has its own profile or provider config, choose the lane there first:

• plugin/provider setting such as space = "Research Agent"
• a derived mapping such as spaceTemplate = "agent-${AGENT_NAME}"
• an exact identity map such as space_by_identity = {"research":"Research Agent"}

Use NMEM_SPACE="Research Agent" only for CLI-first hosts or runtimes that do not expose a better config surface. HTTP- or MCP-based integrations should pass space_id explicitly when their host/runtime can do so. The storage boundary is still one hidden shared key, but humans and agents should normally work with the space name instead. Legacy NMEM_SPACE_ID still works for older setups.

For agent harnesses, the rule is simple:

• If the host can only promise one lane per process or profile, support one fixed ambient space.
• If the host exposes a stable identity or workspace signal, support a derived mapping (spaceTemplate or exact identity mapping).
• If the host does not expose identity cleanly, do not fake per-agent routing.

Use one ambient space only when the host already has a real lane, such as one AI Identity, one project, or one workspace.

What the space profile means is the same everywhere:

• When this space searches decides how far automatic recall expands before the agent starts answering.
• Also search these spaces adds reusable context lanes for retrieval only. It does not move or merge records.
• Agent guidance is read by AI Now and built-in/background agents working in that lane. It changes retrieval and explanation style, not storage.

• Documentation
• Blog
• Report a Bug
• Request a Feature
• hello@nowledge-labs.ai