Merge branch 'claude/add-memory-layer-De82O' into extension/claude-mem

This commit is contained in:
Alex Newman 2026-01-28 20:41:50 -05:00
commit cc60285f49
10 changed files with 1567 additions and 1 deletions

View File

@ -0,0 +1,390 @@
# Clawdbot + Claude-Mem Setup Guide
Complete setup guide for integrating Clawdbot with Claude-Mem persistent memory.
---
## Overview
This guide walks you through:
1. Installing Clawdbot (AI messaging gateway)
2. Installing Claude-Mem (persistent memory system)
3. Connecting them via the `memory-claudemem` plugin
**What you get:**
- Clawdbot routes messages between you and AI across multiple channels (Telegram, Discord, WhatsApp, etc.)
- Claude-Mem captures observations from every tool call and injects relevant context into future sessions
- Progressive disclosure memory tools (`memory_search`, `memory_observations`)
---
## Prerequisites
- **Node.js 22+** (required for both Clawdbot and Claude-Mem)
- **macOS, Linux, or Windows (WSL2)**
- **Git** (for installing from source)
Check your Node version:
```bash
node -v
# Should show v22.x.x or higher
```
---
## Part 1: Install Clawdbot
### Option A: Quick Install (Recommended)
```bash
curl -fsSL https://clawd.bot/install.sh | bash
```
Windows (PowerShell):
```powershell
iwr -useb https://clawd.bot/install.ps1 | iex
```
This:
- Installs Node 22+ if needed
- Installs `clawdbot` globally via npm
- Runs the onboarding wizard
### Option B: Manual Install
If you already have Node 22+:
```bash
npm install -g clawdbot@latest
```
Then run onboarding:
```bash
clawdbot onboard --install-daemon
```
### Verify Installation
```bash
clawdbot --version
clawdbot doctor
```
### Common Issue: "clawdbot not found"
If your shell can't find `clawdbot`, add npm's global bin to your PATH:
```bash
# Add to ~/.zshrc or ~/.bashrc
export PATH="$(npm prefix -g)/bin:$PATH"
```
Then restart your terminal or run `source ~/.zshrc`.
---
## Part 2: Install Claude-Mem
Claude-Mem is a separate system that provides persistent memory. You need to install it independently.
### Option A: Via Claude Code Plugin (if using Claude Code)
If you use Claude Code (Anthropic's CLI), install claude-mem as a plugin:
```bash
# In a Claude Code session:
/plugin marketplace add thedotmack/claude-mem
/plugin install claude-mem
```
Restart Claude Code. The worker starts automatically.
### Option B: Manual Installation (Standalone)
Clone the repository:
```bash
cd ~/Scripts # or wherever you keep projects
git clone https://github.com/thedotmack/claude-mem.git
cd claude-mem
```
Install dependencies and build:
```bash
npm install
npm run build
```
Start the worker:
```bash
npm run worker
# Or run in background:
nohup npm run worker > /tmp/claude-mem-worker.log 2>&1 &
```
### Verify Claude-Mem is Running
The worker runs on port 37777 by default.
```bash
curl http://localhost:37777/api/health
# Should return: OK or {"status":"ok"}
```
Open the web viewer:
```bash
open http://localhost:37777
```
---
## Part 3: Enable the Memory-Claudemem Plugin
The `memory-claudemem` plugin connects Clawdbot to the Claude-Mem worker.
### Step 1: Disable Built-in Memory (Optional but Recommended)
Clawdbot has its own memory system. To avoid conflicts, disable it:
```bash
clawdbot config set agents.defaults.memorySearch.enabled false
```
Verify:
```bash
clawdbot config get agents.defaults.memorySearch.enabled
# Should return: false
```
### Step 2: Enable the Claude-Mem Plugin
The plugin is bundled in the `extensions/memory-claudemem` directory. Enable it:
```bash
clawdbot config set plugins.memory-claudemem.workerUrl "http://localhost:37777"
```
Optional: Adjust timeout (default 10 seconds):
```bash
clawdbot config set plugins.memory-claudemem.workerTimeout 15000
```
### Step 3: Verify Plugin Status
Check the worker connection:
```bash
clawdbot claude-mem status
```
Expected output:
```
✓ Worker running at http://localhost:37777
```
If you see `✗ Worker not responding`, make sure Claude-Mem is running (see Part 2).
---
## Part 4: Test the Integration
### Test 1: Search Memories
```bash
clawdbot claude-mem search "authentication"
```
This queries Claude-Mem for observations matching "authentication".
### Test 2: Run an Agent with Memory
Send a test message through Clawdbot:
```bash
clawdbot agent --message "List files in the current directory"
```
This should:
1. Execute the task
2. Record the observation to Claude-Mem (fire-and-forget)
3. Future queries will have this context injected
### Test 3: Verify Observation Recording
Open the Claude-Mem web viewer:
```bash
open http://localhost:37777
```
You should see new observations for tool calls made by Clawdbot.
---
## Configuration Reference
### Full Config Example
Edit `~/.clawdbot/config.yml`:
```yaml
# Disable built-in memory (use claude-mem instead)
agents:
defaults:
memorySearch:
enabled: false
# Enable claude-mem plugin
plugins:
memory-claudemem:
workerUrl: http://localhost:37777
workerTimeout: 10000
```
### Rollback to Built-in Memory
If you want to switch back to Clawdbot's built-in memory:
```yaml
agents:
defaults:
memorySearch:
enabled: true
plugins:
memory-claudemem:
enabled: false
```
---
## Available Tools
When the plugin is active, these tools are available to the AI agent:
| Tool | Description |
|------|-------------|
| `memory_search` | Search past observations. Returns compact results with IDs. |
| `memory_observations` | Get full details for specific observation IDs. |
### 3-Layer Workflow
1. **Search** (`memory_search`): Get index with IDs (~50-100 tokens/result)
2. **Filter**: Review results, identify relevant IDs
3. **Fetch** (`memory_observations`): Get full details (~500-1000 tokens/result)
This pattern saves ~10x tokens compared to fetching everything.
---
## CLI Commands
| Command | Description |
|---------|-------------|
| `clawdbot claude-mem status` | Check if the worker is responding |
| `clawdbot claude-mem search <query>` | Search memories (JSON output) |
Options for search:
```bash
clawdbot claude-mem search "authentication" --limit 20
```
---
## Troubleshooting
### Worker Not Responding
1. Check if Claude-Mem is running:
```bash
curl http://localhost:37777/api/health
```
2. If not running, start it:
```bash
cd ~/Scripts/claude-mem
npm run worker
```
3. Check logs:
```bash
tail -f /tmp/claude-mem-worker.log
```
### Plugin Not Loading
1. Check plugin list:
```bash
clawdbot plugins list
```
2. Look for `memory-claudemem` in the output.
3. If missing, check the config:
```bash
clawdbot config get plugins.memory-claudemem
```
### No Observations Being Recorded
1. Enable debug logging:
```bash
clawdbot config set logging.level debug
```
2. Run a command and check logs:
```bash
clawdbot agent --message "test"
tail -f ~/.clawdbot/logs/gateway.log | grep claude-mem
```
### Context Not Being Injected
1. Check that there are observations in Claude-Mem (via web UI at http://localhost:37777)
2. Ensure the prompt is longer than 5 characters (short prompts skip injection)
3. Check logs for "injecting X memories into context"
---
## Architecture
```
┌──────────────────┐ ┌──────────────────┐
│ Clawdbot │ │ Claude-Mem │
│ Gateway │ │ Worker │
├──────────────────┤ ├──────────────────┤
│ │ │ │
│ before_agent │────▶│ context inject │
│ after_tool_call │────▶│ POST /observe │
│ │ │ │
│ memory_search │────▶│ GET /search │
│ memory_observations──▶│ POST /batch │
│ │ │ │
└──────────────────┘ └──────────────────┘
│ │
└────────────────────────┘
http://localhost:37777
```
**Hooks:**
- `before_agent_start`: Injects relevant memories as context
- `after_tool_call`: Records observations (fire-and-forget)
**Tools:**
- `memory_search`: Queries Claude-Mem's search API
- `memory_observations`: Fetches full observation details
---
## Next Steps
- **Set up messaging channels**: `clawdbot channels add telegram` (or discord, slack, etc.)
- **Configure the AI model**: `clawdbot config set agents.defaults.model claude-sonnet-4-20250514`
- **Start the gateway**: `clawdbot gateway run`
- **Open the dashboard**: `clawdbot dashboard`
For more information:
- Clawdbot docs: https://docs.clawd.bot
- Claude-Mem docs: https://docs.claude-mem.ai

View File

@ -0,0 +1,619 @@
# Claude-Mem Integration Plan
Integrate `thedotmack/claude-mem` as the primary observation agent and memory layer for Clawdbot.
---
## Phase 0: Documentation Discovery
### Allowed APIs (Clawdbot Plugin System)
**Source:** `/home/user/clawdbot/src/plugins/types.ts:235-274`
| API | Signature | Use Case |
|-----|-----------|----------|
| `api.registerTool` | `(tool: AnyAgentTool \| ClawdbotPluginToolFactory, opts?: ClawdbotPluginToolOptions) => void` | Register memory tools |
| `api.on` | `<K extends PluginHookName>(hookName: K, handler: PluginHookHandlerMap[K]) => void` | Register lifecycle hooks |
| `api.registerCli` | `(registrar: ClawdbotPluginCliRegistrar, opts?: { commands?: string[] }) => void` | Add CLI commands |
| `api.registerService` | `(service: ClawdbotPluginService) => void` | Health check service |
| `api.logger` | `PluginLogger` | Logging (`info`, `warn`, `debug`) |
| `api.pluginConfig` | `Record<string, unknown>` | Plugin configuration |
### Available Hooks
**Source:** `/home/user/clawdbot/src/plugins/types.ts:289-303`
| Hook | Execution | Clawdbot → Claude-Mem |
|------|-----------|----------------------|
| `session_start` | Parallel (fire-and-forget) | → `SessionStart` |
| `before_agent_start` | Sequential (can modify) | → `UserPromptSubmit` + context injection |
| `after_tool_call` | Parallel (fire-and-forget) | → `PostToolUse` |
| `agent_end` | Parallel (fire-and-forget) | → `Stop` |
| `session_end` | Parallel (fire-and-forget) | → `SessionEnd` |
### Allowed APIs (Claude-Mem Worker)
**Source:** GitHub Issue #348, Platform Integration Guide
| Endpoint | Method | Request | Response |
|----------|--------|---------|----------|
| `/api/health` | GET | - | 200 OK |
| `/api/sessions/observations` | POST | `{ claudeSessionId, tool_name, tool_input, tool_response, cwd? }` | `{ id }` |
| `/api/search` | GET | `?query=<q>&type=observations&format=index&limit=<n>` | `SearchResult[]` |
| `/api/observations/batch` | POST | `{ ids: number[], orderBy?, limit? }` | `Observation[]` |
**Note:** `created_at_epoch` is in MILLISECONDS (not seconds).
### Memory Disable Mechanism
**Source:** `/home/user/clawdbot/src/agents/memory-search.ts:276`
```typescript
// Existing kill switch - returns null when disabled
if (!resolved.enabled) return null;
```
**Config location:** `agents.defaults.memorySearch.enabled` (default: `true`)
### Copy-Ready Patterns
| Pattern | Source File | Lines |
|---------|-------------|-------|
| Minimal plugin structure | `/home/user/clawdbot/extensions/memory-core/index.ts` | 1-36 |
| Tool registration | `/home/user/clawdbot/extensions/memory-lancedb/index.ts` | 238-286 |
| Hook registration (`before_agent_start`) | `/home/user/clawdbot/extensions/memory-lancedb/index.ts` | 468-491 |
| Hook registration (`agent_end`) | `/home/user/clawdbot/extensions/memory-lancedb/index.ts` | 496-569 |
| CLI commands | `/home/user/clawdbot/extensions/memory-lancedb/index.ts` | 418-460 |
| Config schema with `parse` + `uiHints` | `/home/user/clawdbot/extensions/memory-lancedb/config.ts` | 61-114 |
| Plugin manifest | `/home/user/clawdbot/extensions/memory-lancedb/clawdbot.plugin.json` | 1-67 |
### Anti-Patterns to Avoid
1. **Do NOT invent API endpoints** - Only use confirmed endpoints from Phase 0
2. **Do NOT use `Type.Union` in tool schemas** - Use `stringEnum`/`optionalStringEnum` instead
3. **Do NOT use raw `format` property** - Reserved keyword in some validators
4. **Do NOT add dependencies to root package.json** - Plugin deps go in plugin's package.json
5. **Do NOT use `workspace:*` in dependencies** - Use in `devDependencies` or `peerDependencies` only
---
## Phase 1: Disable Built-in Memory
### What to Implement
1. Add `memorySearch.enabled: false` config to disable built-in memory tools
2. User configures this in their config file when using claude-mem
**No code changes required.** The existing kill switch at `src/agents/memory-search.ts:276` already returns `null` when `enabled: false`.
### Documentation References
- Config schema: `/home/user/clawdbot/src/config/types.tools.ts:223` (`enabled?: boolean`)
- Kill switch: `/home/user/clawdbot/src/agents/memory-search.ts:276`
- Tool creation guard: `/home/user/clawdbot/src/agents/tools/memory-tool.ts:32`
### Verification Checklist
```bash
# 1. Set config
clawdbot config set agents.defaults.memorySearch.enabled false
# 2. Verify memory tools are not registered
clawdbot agent --dry-run 2>&1 | grep -c "memory_search"
# Expected: 0
# 3. Verify config is persisted
clawdbot config get agents.defaults.memorySearch.enabled
# Expected: false
```
### Anti-Pattern Guards
- Do NOT modify `resolveMemorySearchConfig()` - existing logic sufficient
- Do NOT add new config keys - reuse existing `enabled` flag
---
## Phase 2: Create Plugin Structure
### What to Implement
Create `extensions/memory-claudemem/` by copying structure from `extensions/memory-lancedb/`.
### Documentation References
- Copy from: `/home/user/clawdbot/extensions/memory-lancedb/` (directory structure)
- Manifest spec: `/home/user/clawdbot/docs/plugins/manifest.md:16-42`
### Files to Create
```
extensions/memory-claudemem/
├── package.json # Copy from memory-lancedb/package.json
├── clawdbot.plugin.json # Copy from memory-lancedb/clawdbot.plugin.json
├── index.ts # Plugin entry point
├── config.ts # Config schema
├── client.ts # HTTP client for worker
└── types.ts # Type definitions
```
### package.json (Copy and Modify)
**Copy from:** `/home/user/clawdbot/extensions/memory-lancedb/package.json`
```json
{
"name": "@clawdbot/memory-claudemem",
"version": "0.0.1",
"type": "module",
"main": "index.ts",
"dependencies": {},
"devDependencies": {
"clawdbot": "workspace:*"
},
"peerDependencies": {
"clawdbot": "*"
}
}
```
### clawdbot.plugin.json (Copy and Modify)
**Copy from:** `/home/user/clawdbot/extensions/memory-lancedb/clawdbot.plugin.json:1-20`
```json
{
"id": "memory-claudemem",
"name": "Memory (Claude-Mem)",
"description": "Real-time observation and memory via claude-mem worker",
"kind": "memory",
"version": "0.0.1",
"configSchema": {
"type": "object",
"additionalProperties": false,
"properties": {
"workerUrl": { "type": "string", "default": "http://localhost:37777" },
"workerTimeout": { "type": "number", "default": 10000 }
}
}
}
```
### Verification Checklist
```bash
# 1. Verify plugin loads
cd extensions/memory-claudemem && pnpm install
# 2. Verify manifest is valid JSON
cat clawdbot.plugin.json | jq .
# 3. Verify plugin appears in list
clawdbot plugins list | grep memory-claudemem
```
### Anti-Pattern Guards
- Do NOT add `workspace:*` to `dependencies` (breaks `npm install`)
- Do NOT add plugin deps to root `package.json`
---
## Phase 3: HTTP Client
### What to Implement
Create `client.ts` with methods for claude-mem worker API.
### Documentation References
- Observation endpoint: `POST /api/sessions/observations` (GitHub Issue #348)
- Search endpoint: `GET /api/search?query=<q>&format=index`
- Batch fetch: `POST /api/observations/batch` (v7.3.0+)
- Health: `GET /api/health`
### client.ts
```typescript
// Copy timeout + AbortController pattern from:
// /home/user/clawdbot/src/infra/fetch-with-timeout.ts (if exists)
// Otherwise use standard fetch with AbortController
export class ClaudeMemClient {
constructor(private baseUrl: string, private timeout: number) {}
async ping(): Promise<boolean> {
try {
const res = await fetch(`${this.baseUrl}/api/health`, {
signal: AbortSignal.timeout(this.timeout)
});
return res.ok;
} catch {
return false;
}
}
async observe(sessionId: string, toolName: string, toolInput: unknown, toolResponse: unknown): Promise<void> {
await fetch(`${this.baseUrl}/api/sessions/observations`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
claudeSessionId: sessionId,
tool_name: toolName,
tool_input: toolInput,
tool_response: toolResponse
}),
signal: AbortSignal.timeout(this.timeout)
});
}
async search(query: string, limit = 10): Promise<SearchResult[]> {
const params = new URLSearchParams({
query,
type: "observations",
format: "index",
limit: String(limit)
});
const res = await fetch(`${this.baseUrl}/api/search?${params}`, {
signal: AbortSignal.timeout(this.timeout)
});
return res.json();
}
async getObservations(ids: number[]): Promise<Observation[]> {
const res = await fetch(`${this.baseUrl}/api/observations/batch`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ ids }),
signal: AbortSignal.timeout(this.timeout)
});
return res.json();
}
}
```
### Verification Checklist
```bash
# 1. Start claude-mem worker
claude-mem worker &
# 2. Test health endpoint
curl http://localhost:37777/api/health
# Expected: 200 OK
# 3. Test search endpoint
curl "http://localhost:37777/api/search?query=test&format=index&limit=5"
# Expected: JSON array
```
### Anti-Pattern Guards
- Do NOT invent endpoints like `/api/observe` - use documented `/api/sessions/observations`
- Do NOT assume `created_at_epoch` is in seconds - it's MILLISECONDS
- Do NOT skip error handling - worker may be offline
---
## Phase 4: Hook Registration
### What to Implement
Register hooks that forward events to claude-mem worker.
### Documentation References
- Hook handler signature: `/home/user/clawdbot/src/plugins/types.ts:466-520`
- Example `before_agent_start`: `/home/user/clawdbot/extensions/memory-lancedb/index.ts:468-491`
- Example `agent_end`: `/home/user/clawdbot/extensions/memory-lancedb/index.ts:496-569`
### Hook: after_tool_call → PostToolUse
**Copy pattern from:** `/home/user/clawdbot/extensions/memory-lancedb/index.ts:496-520`
```typescript
// In index.ts register() function
api.on("after_tool_call", async (event, ctx) => {
// Skip our own tools to prevent recursion
if (event.toolName.startsWith("memory_")) return;
try {
await client.observe(
ctx.sessionId,
event.toolName,
event.params,
event.result
);
} catch (err) {
api.logger.warn?.(`claude-mem: observation failed: ${err}`);
}
});
```
### Hook: before_agent_start → Context Injection
**Copy pattern from:** `/home/user/clawdbot/extensions/memory-lancedb/index.ts:468-491`
```typescript
api.on("before_agent_start", async (event, ctx) => {
try {
const results = await client.search(event.prompt, 5);
if (results.length === 0) return;
const context = results
.map(r => `- [#${r.id}] ${r.title}: ${r.snippet}`)
.join("\n");
return {
prependContext: `<claude-mem-context>\n${context}\n</claude-mem-context>`
};
} catch (err) {
api.logger.warn?.(`claude-mem: context injection failed: ${err}`);
}
});
```
### Verification Checklist
```bash
# 1. Enable debug logging
clawdbot config set plugins.memory-claudemem.debug true
# 2. Run a test message
clawdbot message send "test message"
# 3. Check logs for observation
tail -f ~/.clawdbot/logs/gateway.log | grep claude-mem
# 4. Verify in claude-mem UI
open http://localhost:37777
```
### Anti-Pattern Guards
- Do NOT observe `memory_*` tools (causes recursion)
- Do NOT block on observation failures (use fire-and-forget pattern for void hooks)
- Do NOT modify event object for void hooks (only `before_agent_start` can return modifications)
---
## Phase 5: Progressive Disclosure Tools
### What to Implement
Register three tools following claude-mem's 3-layer retrieval pattern.
### Documentation References
- Tool registration: `/home/user/clawdbot/extensions/memory-lancedb/index.ts:238-286`
- Tool factory pattern: `/home/user/clawdbot/src/plugins/types.ts:69-77`
- Tool schema guardrails: Use `Type.Object` with `Type.String`, `Type.Optional`, `Type.Number`
### Layer 1: memory_search (~50-100 tokens per result)
**Copy pattern from:** `/home/user/clawdbot/extensions/memory-lancedb/index.ts:238-270`
```typescript
api.registerTool({
name: "memory_search",
label: "Memory Search",
description: "Search past observations. Returns compact results with IDs. Use memory_observations for full details.",
parameters: Type.Object({
query: Type.String({ description: "Natural language search query" }),
limit: Type.Optional(Type.Number({ description: "Max results (default: 10)" }))
}),
async execute(_toolCallId, params) {
const results = await client.search(params.query, params.limit ?? 10);
const text = results.length === 0
? "No memories found."
: results.map((r, i) => `${i + 1}. [#${r.id}] ${r.title}`).join("\n");
return { content: [{ type: "text", text }] };
}
}, { name: "memory_search" });
```
### Layer 3: memory_observations (~500-1000 tokens per result)
```typescript
api.registerTool({
name: "memory_observations",
label: "Memory Observations",
description: "Get full details for specific observation IDs. Use after memory_search to filter.",
parameters: Type.Object({
ids: Type.Array(Type.Number(), { description: "Observation IDs from memory_search" })
}),
async execute(_toolCallId, params) {
const observations = await client.getObservations(params.ids);
const text = observations.map(o =>
`## #${o.id}\n${o.narrative}\n\nFiles: ${o.files_modified?.join(", ") || "none"}`
).join("\n\n---\n\n");
return { content: [{ type: "text", text }] };
}
}, { name: "memory_observations" });
```
### Verification Checklist
```bash
# 1. Verify tools are registered
clawdbot tools list | grep memory_
# 2. Test search tool
clawdbot agent --message "Use memory_search to find recent file edits"
# 3. Test observations tool
clawdbot agent --message "Use memory_observations with IDs [1, 2, 3]"
```
### Anti-Pattern Guards
- Do NOT use `Type.Union` in parameters - not allowed by schema guardrails
- Do NOT use raw `format` property name - reserved keyword
- Do NOT return tool results without `content: [{ type: "text", text }]` structure
---
## Phase 6: CLI Commands
### What to Implement
Add CLI commands for worker status and manual search.
### Documentation References
- CLI registration: `/home/user/clawdbot/extensions/memory-lancedb/index.ts:418-460`
- Commander.js pattern: `/home/user/clawdbot/src/plugins/types.ts:193-200`
### Implementation
**Copy pattern from:** `/home/user/clawdbot/extensions/memory-lancedb/index.ts:418-460`
```typescript
api.registerCli(({ program }) => {
const mem = program
.command("claude-mem")
.description("Claude-mem integration commands");
mem.command("status")
.description("Check claude-mem worker status")
.action(async () => {
const alive = await client.ping();
console.log(alive
? `✓ Worker running at ${cfg.workerUrl}`
: `✗ Worker not responding at ${cfg.workerUrl}`);
});
mem.command("search")
.description("Search memories")
.argument("<query>", "Search query")
.action(async (query) => {
const results = await client.search(query);
console.log(JSON.stringify(results, null, 2));
});
}, { commands: ["claude-mem"] });
```
### Verification Checklist
```bash
# 1. Verify command appears in help
clawdbot claude-mem --help
# 2. Test status command
clawdbot claude-mem status
# 3. Test search command
clawdbot claude-mem search "authentication"
```
### Anti-Pattern Guards
- Do NOT use reserved command names (help, send, config, status, etc.) - see `/home/user/clawdbot/src/plugins/commands.ts:33-69`
---
## Phase 7: Verification
### Implementation Verification
1. **Memory disabled when configured:**
```bash
clawdbot config set agents.defaults.memorySearch.enabled false
clawdbot tools list | grep -v memory_claudemem | grep memory_
# Expected: no results (built-in memory tools gone)
```
2. **Plugin loads successfully:**
```bash
clawdbot plugins list | grep memory-claudemem
# Expected: memory-claudemem (enabled)
```
3. **Worker connection verified:**
```bash
clawdbot claude-mem status
# Expected: ✓ Worker running at http://localhost:37777
```
4. **Observations recorded:**
```bash
# Run a command
clawdbot agent --message "List files in current directory"
# Check worker UI
open http://localhost:37777
# Expected: New observation for "bash" tool
```
5. **Context injection works:**
```bash
# First, create some observations
clawdbot agent --message "Create a file called test.txt"
# Then search should inject context
DEBUG=claude-mem clawdbot agent --message "What files did I create recently?"
# Expected: Logs show context injection
```
### Anti-Pattern Check
```bash
# Grep for known bad patterns
grep -r "Type\.Union" extensions/memory-claudemem/
# Expected: no results
grep -r "format:" extensions/memory-claudemem/*.ts
# Expected: no results (or only in non-schema contexts)
grep -r "/api/observe" extensions/memory-claudemem/
# Expected: no results (should use /api/sessions/observations)
```
### Test Suite
```bash
cd extensions/memory-claudemem
pnpm test
# Expected: All tests pass
pnpm lint
# Expected: No lint errors
pnpm build
# Expected: Builds without type errors
```
---
## Configuration Summary
### User Config (to use claude-mem)
```yaml
# ~/.clawdbot/config.yml
# Disable built-in memory
agents:
defaults:
memorySearch:
enabled: false
# Enable claude-mem plugin
plugins:
memory-claudemem:
workerUrl: http://localhost:37777
workerTimeout: 10000
```
### Rollback (to restore built-in memory)
```yaml
agents:
defaults:
memorySearch:
enabled: true
plugins:
memory-claudemem:
enabled: false
```

View File

@ -0,0 +1,27 @@
{
"id": "memory-claudemem",
"name": "Memory (Claude-Mem)",
"description": "Real-time observation and memory via claude-mem worker",
"kind": "memory",
"version": "0.0.1",
"uiHints": {
"workerUrl": {
"label": "Worker URL",
"placeholder": "http://localhost:37777",
"help": "URL of the claude-mem worker"
},
"workerTimeout": {
"label": "Timeout (ms)",
"placeholder": "10000",
"advanced": true
}
},
"configSchema": {
"type": "object",
"additionalProperties": false,
"properties": {
"workerUrl": { "type": "string", "default": "http://localhost:37777" },
"workerTimeout": { "type": "number", "default": 10000 }
}
}
}

View File

@ -0,0 +1,143 @@
import type { SearchResult, Observation } from "./types.js";
export class ClaudeMemClient {
constructor(
private readonly baseUrl: string,
private readonly timeout: number,
) {}
/**
* Check if the claude-mem worker is healthy.
* Returns true if the worker responds to health check, false otherwise.
*/
async ping(): Promise<boolean> {
try {
const response = await fetch(`${this.baseUrl}/api/health`, {
method: "GET",
signal: AbortSignal.timeout(this.timeout),
});
return response.ok;
} catch {
return false;
}
}
/**
* Record an observation to the claude-mem worker.
* Fire-and-forget pattern: errors are logged but do not block.
*/
async observe(
claudeSessionId: string,
toolName: string,
toolInput: unknown,
toolResponse: unknown,
cwd?: string,
): Promise<void> {
try {
const body = {
claudeSessionId,
tool_name: toolName,
tool_input: toolInput,
tool_response: toolResponse,
...(cwd && { cwd }),
};
await fetch(`${this.baseUrl}/api/sessions/observations`, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify(body),
signal: AbortSignal.timeout(this.timeout),
});
// Fire-and-forget: we don't check response status or throw on errors
} catch {
// Silently ignore errors - worker may be offline
// This is intentional: observations should never block the main flow
}
}
/**
* Search observations by query.
* Returns an array of search results with id, title, snippet, and score.
*/
async search(query: string, limit = 10): Promise<SearchResult[]> {
try {
const searchParams = new URLSearchParams({
query,
type: "observations",
format: "index",
limit: String(limit),
});
const response = await fetch(
`${this.baseUrl}/api/search?${searchParams.toString()}`,
{
method: "GET",
signal: AbortSignal.timeout(this.timeout),
},
);
if (!response.ok) {
return [];
}
const data = await response.json();
// The API returns results in an array format
if (Array.isArray(data)) {
return data as SearchResult[];
}
// Handle wrapped response format { results: [...] }
if (data && Array.isArray(data.results)) {
return data.results as SearchResult[];
}
return [];
} catch {
return [];
}
}
/**
* Fetch full observation details by IDs.
* Returns observations with narrative, files_modified, tool_name, tool_input, tool_response, and created_at_epoch.
*/
async getObservations(ids: number[]): Promise<Observation[]> {
if (ids.length === 0) {
return [];
}
try {
const response = await fetch(`${this.baseUrl}/api/observations/batch`, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify({ ids }),
signal: AbortSignal.timeout(this.timeout),
});
if (!response.ok) {
return [];
}
const data = await response.json();
// Handle direct array response
if (Array.isArray(data)) {
return data as Observation[];
}
// Handle wrapped response format { observations: [...] }
if (data && Array.isArray(data.observations)) {
return data.observations as Observation[];
}
return [];
} catch {
return [];
}
}
}

View File

@ -0,0 +1,49 @@
export type ClaudeMemConfig = {
workerUrl: string;
workerTimeout: number;
};
const DEFAULT_WORKER_URL = "http://localhost:37777";
const DEFAULT_TIMEOUT = 10000;
function assertAllowedKeys(
value: Record<string, unknown>,
allowed: string[],
label: string,
) {
const unknown = Object.keys(value).filter((key) => !allowed.includes(key));
if (unknown.length === 0) return;
throw new Error(`${label} has unknown keys: ${unknown.join(", ")}`);
}
export const claudeMemConfigSchema = {
parse(value: unknown): ClaudeMemConfig {
if (!value || typeof value !== "object" || Array.isArray(value)) {
// Return defaults if no config provided
return {
workerUrl: DEFAULT_WORKER_URL,
workerTimeout: DEFAULT_TIMEOUT,
};
}
const cfg = value as Record<string, unknown>;
assertAllowedKeys(cfg, ["workerUrl", "workerTimeout"], "claude-mem config");
return {
workerUrl: typeof cfg.workerUrl === "string" ? cfg.workerUrl : DEFAULT_WORKER_URL,
workerTimeout: typeof cfg.workerTimeout === "number" ? cfg.workerTimeout : DEFAULT_TIMEOUT,
};
},
uiHints: {
workerUrl: {
label: "Worker URL",
placeholder: DEFAULT_WORKER_URL,
help: "URL of the claude-mem worker",
},
workerTimeout: {
label: "Timeout (ms)",
placeholder: String(DEFAULT_TIMEOUT),
advanced: true,
help: "Request timeout in milliseconds",
},
},
};

View File

@ -0,0 +1,195 @@
import type { ClawdbotPluginApi } from "clawdbot/plugin-sdk";
import { Type } from "@sinclair/typebox";
import { claudeMemConfigSchema } from "./config.js";
import { ClaudeMemClient } from "./client.js";
const claudeMemPlugin = {
id: "memory-claudemem",
name: "Memory (Claude-Mem)",
description: "Real-time observation and memory via claude-mem worker",
kind: "memory" as const,
configSchema: claudeMemConfigSchema,
register(api: ClawdbotPluginApi) {
const cfg = claudeMemConfigSchema.parse(api.pluginConfig);
const client = new ClaudeMemClient(cfg.workerUrl, cfg.workerTimeout);
api.logger.info(
`memory-claudemem: plugin registered (worker: ${cfg.workerUrl})`,
);
// Hook: after_tool_call → Observe tool calls (fire-and-forget)
api.on("after_tool_call", async (event, ctx) => {
// Skip memory tools to prevent recursion
if (event.toolName.startsWith("memory_")) return;
try {
// Fire-and-forget: don't await, let it run in parallel
// Use sessionKey as the session identifier (may be undefined)
client.observe(
ctx.sessionKey ?? "unknown",
event.toolName,
event.params,
event.result,
);
} catch (err) {
api.logger.warn?.(`memory-claudemem: observation failed: ${err}`);
}
});
// Hook: before_agent_start → Context injection from memory search
api.on("before_agent_start", async (event) => {
// Skip if prompt is empty or too short
if (!event.prompt || event.prompt.length < 5) return;
try {
const results = await client.search(event.prompt, 5);
if (results.length === 0) return;
const memoryContext = results
.map((r) => `- [#${r.id}] ${r.title}: ${r.snippet}`)
.join("\n");
api.logger.info?.(
`memory-claudemem: injecting ${results.length} memories into context`,
);
return {
prependContext: `<claude-mem-context>\nThe following memories may be relevant:\n${memoryContext}\n</claude-mem-context>`,
};
} catch (err) {
api.logger.warn?.(`memory-claudemem: context injection failed: ${err}`);
}
});
// Phase 5: Tool registration
// memory_search - Layer 1: compact results (~50-100 tokens per result)
api.registerTool(
{
name: "memory_search",
label: "Memory Search",
description:
"Search past observations. Returns compact results with IDs. Use memory_observations for full details.",
parameters: Type.Object({
query: Type.String({ description: "Search query" }),
limit: Type.Optional(
Type.Number({ description: "Max results (default: 10)" }),
),
}),
async execute(_toolCallId, params) {
const { query, limit = 10 } = params as {
query: string;
limit?: number;
};
const results = await client.search(query, limit);
if (results.length === 0) {
return {
content: [{ type: "text", text: "No relevant memories found." }],
};
}
const text = results.map((r) => `[#${r.id}] ${r.title}`).join("\n");
return {
content: [
{
type: "text",
text: `Found ${results.length} memories:\n\n${text}`,
},
],
};
},
},
{ name: "memory_search" },
);
// memory_observations - Layer 3: full details (~500-1000 tokens per result)
api.registerTool(
{
name: "memory_observations",
label: "Memory Observations",
description:
"Get full details for specific observation IDs. Use after memory_search to filter.",
parameters: Type.Object({
ids: Type.Array(Type.Number(), {
description: "Array of observation IDs to fetch",
}),
}),
async execute(_toolCallId, params) {
const { ids } = params as { ids: number[] };
if (ids.length === 0) {
return {
content: [{ type: "text", text: "No observation IDs provided." }],
};
}
const observations = await client.getObservations(ids);
if (observations.length === 0) {
return {
content: [
{
type: "text",
text: "No observations found for the given IDs.",
},
],
};
}
const text = observations
.map((obs) => {
const filesSection =
obs.files_modified && obs.files_modified.length > 0
? `\n\nFiles: ${obs.files_modified.join(", ")}`
: "";
return `## #${obs.id}\n${obs.narrative}${filesSection}`;
})
.join("\n\n---\n\n");
return {
content: [{ type: "text", text }],
};
},
},
{ name: "memory_observations" },
);
// Phase 6: CLI registration
api.registerCli(
({ program }) => {
const claudeMem = program
.command("claude-mem")
.description("Claude-mem memory plugin commands");
claudeMem
.command("status")
.description("Check if the claude-mem worker is responding")
.action(async () => {
const isHealthy = await client.ping();
if (isHealthy) {
console.log(`✓ Worker running at ${cfg.workerUrl}`);
} else {
console.log(`✗ Worker not responding at ${cfg.workerUrl}`);
process.exitCode = 1;
}
});
claudeMem
.command("search")
.description("Search memories")
.argument("<query>", "Search query")
.option("--limit <n>", "Max results", "10")
.action(async (query, opts) => {
const results = await client.search(query, parseInt(opts.limit));
console.log(JSON.stringify(results, null, 2));
});
},
{ commands: ["claude-mem"] },
);
},
};
export default claudeMemPlugin;

View File

@ -0,0 +1,13 @@
{
"name": "@clawdbot/memory-claudemem",
"version": "0.0.1",
"type": "module",
"main": "index.ts",
"dependencies": {},
"devDependencies": {
"clawdbot": "workspace:*"
},
"peerDependencies": {
"clawdbot": "*"
}
}

View File

@ -0,0 +1,24 @@
/** Search result from claude-mem worker (index format) */
export type SearchResult = {
id: number;
title: string;
snippet?: string;
score?: number;
};
/** Full observation from claude-mem worker */
export type Observation = {
id: number;
narrative: string;
files_modified?: string[];
tool_name?: string;
tool_input?: unknown;
tool_response?: unknown;
created_at_epoch?: number;
};
/** Plugin configuration */
export type ClaudeMemConfig = {
workerUrl: string;
workerTimeout: number;
};

View File

@ -283,5 +283,10 @@
"apps/macos/.build/**",
"dist/Moltbot.app/**"
]
}
},
"trustedDependencies": [
"@whiskeysockets/baileys",
"node-llama-cpp",
"protobufjs"
]
}

101
pnpm-lock.yaml generated
View File

@ -261,6 +261,13 @@ importers:
node-llama-cpp:
specifier: 3.15.0
version: 3.15.0(typescript@5.9.3)
optionalDependencies:
'@napi-rs/canvas':
specifier: ^0.1.88
version: 0.1.88
node-llama-cpp:
specifier: 3.15.0
version: 3.15.0(typescript@5.9.3)
extensions/bluebubbles: {}
@ -354,11 +361,21 @@ importers:
extensions/mattermost: {}
extensions/memory-claudemem:
devDependencies:
clawdbot:
specifier: workspace:*
version: link:../..
extensions/memory-core:
devDependencies:
moltbot:
specifier: workspace:*
version: link:../..
dependencies:
clawdbot:
specifier: '>=2026.1.24'
version: 2026.1.24(@types/express@5.0.6)(audio-decode@2.2.3)(devtools-protocol@0.0.1561482)(typescript@5.9.3)
extensions/memory-lancedb:
dependencies:
@ -1323,6 +1340,7 @@ packages:
resolution: {integrity: sha512-aYrIoEG24AC+wILCL57Ius/Y4yU+xFHDPKLvmjzzN4byAjzeIGF0TC86S5RBt4Ji+dxS7yIWV5Q/gE5/fybIFQ==}
engines: {node: '>= 18'}
cpu: [x64, arm64]
cpu: [x64, arm64]
os: [darwin, linux, win32]
peerDependencies:
apache-arrow: '>=15.0.0 <=18.1.0'
@ -3219,6 +3237,11 @@ packages:
engines: {node: '>=22.12.0'}
hasBin: true
clawdbot@2026.1.24:
resolution: {integrity: sha512-foszbNXzk743kQBx2Nfc2KNlStZyBAVAYLQJ+KaONh009r8oB1a74kQ8wTnKX+SDNaQ9MvaE9tOisVUi+H3F+Q==}
engines: {node: '>=22.12.0'}
hasBin: true
cli-cursor@5.0.0:
resolution: {integrity: sha512-aCj4O5wKyszjMmDT4tZj93kxyydN/K5zPWSCe6/0AV/AA1pqe5ZBIw0a2ZfPQV7lL5/yb5HsUreJ6UFAF1tEQw==}
engines: {node: '>=18'}
@ -9176,6 +9199,84 @@ snapshots:
- typescript
- utf-8-validate
clawdbot@2026.1.24(@types/express@5.0.6)(audio-decode@2.2.3)(devtools-protocol@0.0.1561482)(typescript@5.9.3):
dependencies:
'@agentclientprotocol/sdk': 0.13.1(zod@4.3.6)
'@aws-sdk/client-bedrock': 3.975.0
'@buape/carbon': 0.14.0(hono@4.11.4)
'@clack/prompts': 0.11.0
'@grammyjs/runner': 2.0.3(grammy@1.39.3)
'@grammyjs/transformer-throttler': 1.2.1(grammy@1.39.3)
'@homebridge/ciao': 1.3.4
'@line/bot-sdk': 10.6.0
'@lydell/node-pty': 1.2.0-beta.3
'@mariozechner/pi-agent-core': 0.49.3(ws@8.19.0)(zod@4.3.6)
'@mariozechner/pi-ai': 0.49.3(ws@8.19.0)(zod@4.3.6)
'@mariozechner/pi-coding-agent': 0.49.3(ws@8.19.0)(zod@4.3.6)
'@mariozechner/pi-tui': 0.49.3
'@mozilla/readability': 0.6.0
'@sinclair/typebox': 0.34.47
'@slack/bolt': 4.6.0(@types/express@5.0.6)
'@slack/web-api': 7.13.0
'@whiskeysockets/baileys': 7.0.0-rc.9(audio-decode@2.2.3)(sharp@0.34.5)
ajv: 8.17.1
body-parser: 2.2.2
chalk: 5.6.2
chokidar: 5.0.0
chromium-bidi: 13.0.1(devtools-protocol@0.0.1561482)
cli-highlight: 2.1.11
commander: 14.0.2
croner: 9.1.0
detect-libc: 2.1.2
discord-api-types: 0.38.37
dotenv: 17.2.3
express: 5.2.1
file-type: 21.3.0
grammy: 1.39.3
hono: 4.11.4
jiti: 2.6.1
json5: 2.2.3
jszip: 3.10.1
linkedom: 0.18.12
long: 5.3.2
markdown-it: 14.1.0
node-edge-tts: 1.2.9
osc-progress: 0.3.0
pdfjs-dist: 5.4.530
playwright-core: 1.58.0
proper-lockfile: 4.1.2
qrcode-terminal: 0.12.0
sharp: 0.34.5
sqlite-vec: 0.1.7-alpha.2
tar: 7.5.4
tslog: 4.10.2
undici: 7.19.0
ws: 8.19.0
yaml: 2.8.2
zod: 4.3.6
optionalDependencies:
'@napi-rs/canvas': 0.1.88
node-llama-cpp: 3.15.0(typescript@5.9.3)
transitivePeerDependencies:
- '@discordjs/opus'
- '@modelcontextprotocol/sdk'
- '@types/express'
- audio-decode
- aws-crt
- bufferutil
- canvas
- debug
- devtools-protocol
- encoding
- ffmpeg-static
- jimp
- link-preview-js
- node-opus
- opusscript
- supports-color
- typescript
- utf-8-validate
cli-cursor@5.0.0:
dependencies:
restore-cursor: 5.1.0