Nowledge Community

Registry

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.

Community Linux packages

The official Linux release channels remain the APT repository, .deb, AppImage, and Docker paths documented at mem.nowledge.co/docs/installation. The packages below are maintained by community contributors for users who prefer their distro's native package manager. Review the package metadata or PKGBUILD before installing, as you would for any community-maintained package.

Target	Package	Maintainer	Install
Fedora / RPM-based Linux	Fedora COPR: abn/nowledge-mem	abn	sudo dnf install dnf-plugins-core sudo dnf copr enable abn/nowledge-mem sudo dnf install nowledge-mem
Fedora / RPM-based headless server	nowledge-mem-server from the same COPR	abn	sudo dnf install nowledge-mem-server
Fedora / RPM-based CLI only	nowledge-mem-cli from the same COPR	abn	sudo dnf install nowledge-mem-cli
Arch Linux desktop app	AUR: nowledge-mem or nowledge-mem-bin	Community-maintained	yay -S nowledge-mem or yay -S nowledge-mem-bin
Arch Linux CLI only	AUR: nmem-cli	czyt	yay -S nmem-cli or paru -S nmem-cli

Integrations

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

Integration	Install	What it does
Universal Agent Plugin	Install this source directory from an Agent Plugins-compatible client	Standards-compatible fallback for clients that support Agent Plugins but do not have a dedicated Nowledge connector yet. Provides skills plus local MCP; does not claim automatic full-thread capture.
Skills	npx skills add nowledge-co/community/nowledge-mem-npx-skills	Reusable workflow package for Context Bundle / Working Memory startup context, routed recall, resumable handoffs, and distillation. Prefer native packages when your tool has one.
Claude Code Plugin	claude plugin marketplace add https://github.com/nowledge-co/community then claude plugin install nowledge-mem@nowledge-community	Claude Code native plugin with hooks for Context Bundle / Working Memory startup context, routed recall, automatic session capture, and pre-compaction transcript save.
Grok Build Plugin	grok plugin install nowledge-co/community#nowledge-mem-claude-code-plugin --trust	Grok loads the shared Claude-compatible package, with Grok-aware Context Bundle startup and nmem t save --from grok session capture.
Grok Bot connector	Add a custom MCP connector in Grok Bot settings	Grok Bot is cloud-hosted: use a public HTTPS Mem endpoint. Its Plugins settings may also expose the shared package and its supported skills/hooks, but automatic full-thread capture is not promised until Grok Bot exposes a documented transcript contract.
Copilot CLI Plugin	copilot plugin marketplace add nowledge-co/community then copilot plugin install nowledge-mem@nowledge-community	GitHub Copilot CLI plugin with startup context guidance, routed recall, incremental session capture, and pre-compaction transcript save.
Droid Plugin	droid plugin marketplace add https://github.com/nowledge-co/community then droid plugin install nowledge-mem@nowledge-community	Factory Droid plugin with Context Bundle / Working Memory startup context, routed recall, distillation, and honest save-handoff semantics.
Gemini CLI	Search Nowledge Mem in the Gemini CLI Extensions Gallery and install	Gemini-native context, bundled MCP, hooks, commands, and skills for Context Bundle / Working Memory startup context, routed recall, real thread save before compression or exit, and handoff summaries.
Google Antigravity Plugin	mkdir -p ~/.gemini/config/plugins && git clone --depth 1 https://github.com/nowledge-co/nowledge-mem-google-antigravity.git ~/.gemini/config/plugins/nowledge-mem	Native Antigravity plugin contributed by abn: Context Bundle / Working Memory startup context, bundled MCP, skills, lifecycle hooks, offline retry, and automatic transcript capture as source=antigravity.
Antigravity Trajectory Extractor	Current sessions: prefer the Google Antigravity plugin or nmem t sync --from antigravity; older cache-only sessions: clone the extractor	Fallback extraction for old Antigravity local caches that do not have the documented transcript.jsonl file.
Windsurf Trajectory Extractor	git clone https://github.com/jijiamoer/windsurf-trajectory-extractor.git	Offline protobuf extraction for Windsurf Cascade conversation history.
Cursor Plugin	Link nowledge-mem-cursor-plugin into ~/.cursor/plugins/local/nowledge-mem-cursor	Cursor-native plugin with session-start context, MCP recall, exact-session transcript capture, manual save-thread, and summary-only save-handoff.
Codex Plugin	codex plugin marketplace add nowledge-co/community --sparse .agents --sparse nowledge-mem-codex-plugin then codex plugin add nowledge-mem@nowledge-community; enable plugins and hooks, run setup, then trust the hooks when Codex asks	Native startup context and memory routing, bundled MCP retrieval and writes, plus Stop-hook capture of real Codex sessions. Coexists cleanly with Codex local Memory.
Raft	Configure per-worker environment variables in Raft runtime config	Install the child runtime connector, set NMEM_AGENT_ID=<agent-slug> per worker, and use Mem skills/MCP inside Raft. Raft-managed Codex rollouts are execution traces and are excluded from Codex Thread capture.
Lody	Configure the child runtime in Lody Agent Config	Runtime launcher setup: install the child runtime connector first; set NMEM_AGENT_ID=<agent-slug> only when that Agent Config represents a stable role.
Multica	Configure the Multica agent MCP settings and custom environment	Multi-agent orchestrator setup: install the child runtime connector first; for Claude Code agents, save the Mem MCP config on the Multica agent; then set NMEM_AGENT_ID=<agent-slug> in custom env.
Cumora	Configure the child runtime plus each teammate persona	AI teammate workspace setup: connect Mem at the runtime/daemon boundary, then use a per-persona Context Bundle instruction unless Cumora exposes per-agent runtime environment variables.
Paseo	Configure the child runtime that Paseo launches	Multi-agent orchestration setup: install the connector for Codex, Claude Code, OpenCode, Pi, or OMP first for automatic new-thread capture; use nmem t sync --from paseo --all-projects --apply to import supported child-session history from Paseo's registry without creating duplicate child threads; set NMEM_AGENT_ID=<agent-slug> only for durable Paseo agent roles.
Cindy	nmem config mcp show --host cindy	Cindy host setup: add Mem as a Cindy MCP server and shared guide/skills for memory behavior. For exact transcript capture, install the dedicated connector for the runtime Cindy launches, such as Codex, Claude Code, Pi, or OMP.
OpenClaw Plugin	openclaw plugins install clawhub:@nowledge/openclaw-nowledge-mem	Full memory lifecycle with memory tools, thread tools, automatic capture, and distillation.
Alma Plugin	Search Nowledge in Alma official Plugin marketplace	Alma-native plugin with startup context, thread-aware recall, structured saves, and optional auto-capture.
Bub Plugin	pip install nowledge-mem-bub	Bub-native plugin: cross-tool knowledge, auto-capture via save_state, startup context, and graph exploration.
Pi Package	pi install npm:nowledge-mem-pi	Pi-native package with startup Context Bundle / Working Memory injection, automatic conversation capture, and bundled memory skills.
OMP Plugin	omp plugin install nowledge-mem-omp	OMP-native plugin with startup Context Bundle / Working Memory injection, automatic conversation capture, and bundled memory skills.
OpenCode Plugin	Add "opencode-nowledge-mem" to opencode.json plugins	Native OpenCode plugin with Context Bundle, Working Memory, search/save tools, idle-event automatic thread capture, pre-compaction flush, handoff, and status.
Amp Plugin	bash nowledge-mem-amp-plugin/scripts/install.sh	Native Amp plugin with Context Bundle, Working Memory, search/save tools, agent.end automatic thread capture, handoff, status, and an Amp skill.
Craft Agent Connector	nmem config mcp show --host craft-agent	Craft workspace source + guide setup: MCP memory tools inside Craft Agent, plus nmem t sync --from craft-agent for real local session import from Craft session.jsonl files.
WorkBuddy Plugin	In WorkBuddy: /plugin marketplace add https://raw.githubusercontent.com/nowledge-co/community/main/.workbuddy-plugin/marketplace.json --name nowledge-community, then /plugin install nowledge-mem@nowledge-community	WorkBuddy-native setup through the CodeBuddy-compatible plugin abstraction: startup context, bundled MCP, slash commands, lifecycle hooks, and real WorkBuddy thread capture through nmem using WorkBuddy's transcript_path.
Devin Plugin	devin plugins install nowledge-co/community#nowledge-mem-devin-plugin	Devin-native MCP, memory skills, lifecycle capture, selected-chain local import, and read-only enterprise Cloud v3 synchronization. Native plugins are currently a Devin closed beta.
CodeBuddy Plugin	codebuddy plugin marketplace add nowledge-co/community then codebuddy plugin install nowledge-mem@nowledge-community	CodeBuddy-native plugin with startup context, bundled MCP, slash commands, lifecycle hooks, and real CodeBuddy thread capture through nmem using CodeBuddy's transcript_path.
Kimi Code Plugin	In Kimi Code, run /plugins install https://github.com/nowledge-co/community/tree/main, then /reload	Kimi-native plugin metadata, session-start skill, native lifecycle hooks, slash commands, user-owned MCP config via nmem config mcp show --host kimi-code, and real Kimi Code thread capture through nmem.
Kimi Work Connector	python3 ~/.cache/nowledge-community/nowledge-mem-kimi-work-connector/scripts/install_kimi_work_plugin.py	Kimi Work desktop connector for its embedded Kimi Code runtime: session-start skill, bundled local MCP, and explicit nmem t sync --from kimi-work session import.
ZCode Plugin	In ZCode, add https://github.com/nowledge-co/zcode-plugin from Settings → Plugins → Create → Add marketplace, then install nowledge-mem-zcode	ZCode-native plugin with bundled MCP, Skills, commands, startup/recall hooks, and Stop-hook transcript capture through nmem. Verify one short session after install because hook support depends on the active ZCode build.
DeepSeek Harness Plugin	dsh plugin --profile web add github:nowledge-co/nowledge-mem-deepseek-harness	Community DSH bundle with native Context Bundle injection, prompt-time recall, Mem MCP tools, and session/event turn-end transcript capture through nmem. The standalone repo is tagged dsh-plugin for DeepSeek Harness ecosystem discovery.
Hermes Agent	bash <(curl -sL https://raw.githubusercontent.com/nowledge-co/community/main/nowledge-mem-hermes/setup.sh)	Native Hermes memory provider with Context Bundle / Working Memory startup context, pre-turn recall, clean nmem_ tools, and session-end transcript capture into Mem threads. MCP remains available as a fallback mode.
Proma Plugin	Manual setup with MCP, hooks, and skills; see Proma guide	Proma desktop agent setup with startup context, Stop-hook thread capture, MCP memory tools, and standard Nowledge Mem skills.
LangGraph Connector	pip install nowledge-mem-langgraph	Python middleware and explicit StateGraph helpers for transient Context Bundle injection, identity-scoped MCP tools, and idempotent top-level Thread capture without replacing LangGraph checkpoints.
Cradle	Enable the bundled Nowledge Mem plugin in Cradle's Plugin Marketplace	Cradle's official adapter provides guided Working Memory, Context Bundle, memory, and thread operations, with optional direct MCP registration. Automatic recall and session capture await Cradle lifecycle hooks.
Arkloop	Settings > Memory > Enable > select Nowledge	Arkloop's official Memory provider injects Working Memory and relevant recall, keeps Arkloop threads in Mem, and can automatically triage and distill completed conversations.
OpticLM	Enable Nowledge Mem from Optic's Extension settings	Optic's official integration uses its own @opticlm/nmem SDK for Mem spaces and thread continuity. Add Mem MCP separately when you also want agent-facing Working Memory and recall tools.
Raycast Extension	Search Nowledge in Raycast Extension Store	Search memories from Raycast launcher.
Claude Desktop	Download from nowled.ge/claude-dxt, double-click .mcpb file	One-click extension for Claude Desktop with memory search, save, and update.
Browser Extension	Install from Chrome Web Store	Side-panel capture for ChatGPT, Claude, Gemini, Perplexity, and other web AI surfaces.
MCP	For tools without a dedicated Nowledge package, use direct MCP.	Standard memory and thread tools exposed through one shared MCP server.

Direct MCP

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
nmem config mcp show --host cindy
nmem config mcp show --host zcode
nmem config mcp show --host deepseek-harness

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

Requirements

• 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

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.

Space behavior by integration

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

Integration	Ambient space today	Best user setup
Claude Code, Grok, Codex, Droid, Pi, Gemini CLI, DeepSeek Harness	Full ambient lane through NMEM_SPACE or per-command --space	Set one NMEM_SPACE only when the whole session truly belongs to one lane. Otherwise stay on Default.
Hermes	Full ambient lane through provider space, space_by_identity, space_template, or fallback NMEM_SPACE	Use space for one stable lane, space_by_identity for a small explicit map, space_template for one lane per Hermes identity.
Alma	Full ambient lane through plugin nowledgeMem.space, plugin nowledgeMem.spaceTemplate, or fallback NMEM_SPACE	Use space for one Alma profile per lane. Use spaceTemplate only when your launcher already exports a trustworthy lane variable.
Bub	Full ambient lane through NMEM_SPACE	Treat Bub as one process-wide lane. If you need separate lanes, run separate Bub processes or profiles.
OpenClaw	Full ambient lane through plugin space, plugin spaceTemplate, or fallback NMEM_SPACE, preserved across CLI memory calls and API-backed thread/feed paths	Use space for one stable profile. Use spaceTemplate only when the launcher already exports the lane signal. Do not fake per-agent routing if the runtime does not expose identity.
OpenCode	Full ambient lane through NMEM_SPACE, preserved across CLI memory calls and HTTP session save	Set one NMEM_SPACE when the OpenCode process belongs to one real lane.
Craft Agent	Partial today	MCP source calls use the configured Mem endpoint. nmem t save/sync --from craft-agent can use --space or NMEM_SPACE for imports; keep one lane only when the Craft workspace represents one real lane.
Cursor	Partial today	sessionStart and handoff flows can follow NMEM_SPACE, but MCP tool calls still need Cursor/runtime support to forward space_id.
Raycast	One fixed lane through Raycast preferences or shared config	Use one named space when that launcher profile always belongs to one lane. Leave it empty to stay on Default.
Browser extension	One fixed lane through extension settings	Use one named space when that browser profile always belongs to one lane. Leave it empty to stay on Default.
Generic MCP-only hosts	Usually default lane only today	Keep using Default unless the host can explicitly pass space_id.

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.

Links

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

---