add changes to docs and test

This commit is contained in:
Hande 2026-01-29 21:55:47 +01:00
parent c71eca5237
commit e8b4540c25
4 changed files with 75 additions and 562 deletions

View File

@ -1,66 +1,37 @@
---
summary: "Cognee knowledge graph memory: setup, Docker config, and usage"
summary: "Cognee memory quick setup and usage"
read_when:
- Setting up Cognee memory provider
- Configuring knowledge graph memory
- Running Cognee with Docker
---
# Cognee Memory Provider
MoltBot supports [**Cognee**](https://www.cognee.ai/) as an optional memory provider. Unlike the default SQLite-based vector memory, Cognee builds a knowledge graph with entity extraction and semantic relationships, providing richer contextual memory for your AI agent.
Moltbot supports [Cognee](https://www.cognee.ai/) - [open source AI memory](https://github.com/topoteretes/cognee) - as an optional memory provider. Cognee builds knowledge graph memory backed by embeddings from any data and can be run locally with Docker. Learn more from [Cognee Documentation](https://docs.cognee.ai/).
## What is Cognee?
## Quickstart with Docker
Cognee is an [open source AI memory](https://github.com/topoteretes/cognee) framework that:
- Ingests unstructured/structured data from 30+ sources
- Extracts entities (people, places, concepts) from documents
- Builds a knowledge graph of relationships backed by embeddings
- Enables semantic search with LLM-powered reasoning
- Supports multiple search modes (e.g., GRAPH_COMPLETION, chunks, summaries)
Learn more at [docs.cognee.ai](https://docs.cognee.ai/).
## Setup Options
### Option 1: Local Docker (Recommended)
Use the repo example compose file:
Run the example compose file:
```bash
docker compose -f examples/cognee-docker-compose.yaml up -d
```
This binds to `127.0.0.1:8000`, so Cognee is only reachable from your machine.
**Verify:**
Verify:
```bash
curl http://localhost:8000/health
```
You can use `http://localhost:8000/docs` to register user and login to get your token.
### Option 2: Cognee Cloud
Use the hosted Cognee service:
1. Sign up at [platform.cognee.ai](https://platform.cognee.ai/)
2. Get your API key from the dashboard
3. Use base URL: `https://cognee--cognee-saas-backend-serve.modal.run`
## Configuration
Moltbot reads `~/.clawdbot/moltbot.json` (JSON5). For local + secure setup, use an env var for the Cognee token and reference it in config.
### Step 1: Put tokens in `~/.clawdbot/.env`
Put the token in `~/.clawdbot/.env`:
```bash
COGNEE_API_KEY="your-cognee-access-token"
CLAWDBOT_GATEWAY_TOKEN="your-random-gateway-token"
```
### Step 2: Configure Cognee in `~/.clawdbot/moltbot.json`
Configure `~/.clawdbot/moltbot.json` (JSON5):
```json5
{
@ -86,195 +57,30 @@ CLAWDBOT_GATEWAY_TOKEN="your-random-gateway-token"
}
```
### Step 3: Start the gateway with env loaded
Start the gateway with env loaded:
```zsh
set -a; source "$HOME/.clawdbot/.env"; set +a
pnpm moltbot gateway --port 18789 --token "$CLAWDBOT_GATEWAY_TOKEN" --verbose
```
## Configuration Options
| Option | Type | Default | Description |
|--------|------|---------|-------------|
| `baseUrl` | string | `http://localhost:8000` | Cognee API endpoint |
| `apiKey` | string | - | Access token (required if Cognee auth is enabled) |
| `datasetName` | string | `"clawdbot"` | Dataset name for organizing memories |
| `searchType` | string | `"GRAPH_COMPLETION"` | Search mode: `GRAPH_COMPLETION`, `chunks`, or `summaries` |
| `maxResults` | number | `6` | Maximum search results returned |
| `autoCognify` | boolean | `true` | Auto-process documents after adding |
| `cognifyBatchSize` | number | `100` | Batch size for processing |
| `timeoutSeconds` | number | `180` | Request timeout in seconds |
## Search Types
Cognee offers these modes. GRAPH_COMPLETION is the default mode.
Learn more from [cognee documentation](https://docs.cognee.ai/core-concepts/main-operations/search)
## Usage
### Memory Files
Cognee indexes `MEMORY.md` in workspace root, `memory/*.md`, and session transcripts when `sources: ["sessions"]` is enabled.
Cognee automatically syncs your memory files:
- `MEMORY.md` or `memory.md` in workspace root
- All `*.md` files in `memory/` directory
### Session Transcripts (Optional)
Enable session memory to index conversation history:
```yaml
agents:
defaults:
memorySearch:
provider: cognee
sources: [memory, sessions] # Include sessions
experimental:
sessionMemory: true
```
### Manual Sync and Status
1. Initial index and status:
```bash
# Manually sync cognee memory for the current agent
pnpm moltbot memory status --index --json
```
## How It Works
2. Memory updates:
1. **Add**: Memory files are sent to Cognee with metadata
2. **Cognify**: Cognee processes documents:
- Extracts entities (people, places, concepts)
- Identifies relationships
- Builds knowledge graph
3. **Search**: Agent queries use semantic search:
- Searches knowledge graph
- Returns relevant insights/chunks/summaries
- Includes metadata and scores
## Comparison: Cognee vs SQLite Memory
| Feature | Cognee | SQLite (Default) |
|---------|--------|------------------|
| **Setup** | Requires Docker/cloud | Built-in, no setup |
| **Offline** | Yes (Optional cloud) | Yes (fully local) |
| **Search** | Knowledge graph + LLM | Vector + BM25 hybrid |
| **Entities** | Extracted automatically | Not available |
| **Relationships** | Yes (graph-based) | No |
| **Speed** | Slower (API calls) | Faster (local DB) |
| **Memory** | Stored locally (Optional cognee configs) | SQLite file |
| **Best for** | Rich context, reasoning | Fast lookup, privacy |
```bash
pnpm moltbot memory status --index --update-cognee --json
```
## Troubleshooting
### Connection Failed
**Error**: `Failed to connect to Cognee at http://localhost:8000`
**Solutions**:
1. Verify Docker is running: `docker ps | grep cognee`
2. Check Cognee logs: `docker logs cognee`
3. Test manually: `curl http://localhost:8000/health`
4. Ensure port 8000 is not blocked
### Slow Performance
**Solutions**:
1. Reduce `maxResults` (try 3-5 instead of 10+)
2. Use `searchType: "chunks"` for faster results
3. Set `autoCognify: false` and cognify manually
4. Check Docker resource limits
### Out of Memory
**Solutions**:
1. Increase Docker memory limit (Docker Desktop settings)
2. Reduce `cognifyBatchSize` (try 50 instead of 100)
3. Process fewer files at once
4. Clear old datasets via Cognee API
### 401 Unauthorized (Cognee auth)
**Cause**: Cognee auth is enabled, but Moltbot is missing/using an invalid `COGNEE_API_KEY`.
**Fix**:
1. Log in to Cognee and get a fresh access token.
2. Update `COGNEE_API_KEY` in `.clawd/.env`.
3. Restart the gateway with env loaded (see Step 3 above).
## Advanced Configuration
### Per-Agent Override
```yaml
agents:
defaults:
memorySearch:
provider: openai # Default for all agents
agents:
research-bot:
memorySearch:
provider: cognee # Override for this agent
cognee:
searchType: GRAPH_COMPLETION
maxResults: 10
```
## Docker Production Tips
### Health Checks
Add health checks to docker-compose.yml:
```yaml
services:
cognee:
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
interval: 30s
timeout: 10s
retries: 3
start_period: 40s
```
### Resource Limits
```yaml
services:
cognee:
deploy:
resources:
limits:
cpus: '2'
memory: 4G
reservations:
cpus: '1'
memory: 2G
```
### Persistent Storage
Mount volumes for persistence:
```yaml
volumes:
- ./cognee_data:/app/cognee/.cognee_system
- ./cognee_logs:/app/logs
```
## Resources
- [Cognee Documentation](https://docs.cognee.ai/)
- [Cognee GitHub](https://github.com/topoteretes/cognee)
- [Clawdbot Memory Guide](/memory)
- [Docker Setup Guide](/install/docker)
## Feedback
Cognee integration is new. Report issues at:
- Clawdbot: [github.com/clawdbot/clawdbot/issues](https://github.com/clawdbot/clawdbot/issues)
- Cognee: [github.com/topoteretes/cognee/issues](https://github.com/topoteretes/cognee/issues)
- Connection test: `curl http://localhost:8000/health`
- Reset cached values that Moltbot reuses: `mv "$HOME/.clawdbot/memory/cognee/main.json" "$HOME/.clawdbot/memory/cognee/main.json.bak"`

View File

@ -35,7 +35,7 @@ agents:
cognifyBatchSize: 100
# Request timeout in seconds
timeoutSeconds: 30
timeoutSeconds: 180
# Enable experimental session memory indexing
experimental:

View File

@ -1,227 +0,0 @@
import { describe, expect, it, vi, beforeEach, afterEach } from "vitest";
import { CogneeClient } from "./cognee-client.js";
import { request } from "undici";
vi.mock("undici", async () => {
const actual = await vi.importActual<typeof import("undici")>("undici");
return {
...actual,
request: vi.fn(),
};
});
describe("CogneeClient", () => {
beforeEach(() => {
vi.clearAllMocks();
});
afterEach(() => {
vi.restoreAllMocks();
});
describe("add", () => {
it("should add data successfully", async () => {
const mockResponse = {
statusCode: 200,
body: {
json: vi.fn().mockResolvedValue({
dataset_id: "test-dataset-id",
dataset_name: "test-dataset",
message: "Data added successfully",
}),
text: vi.fn(),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient({
baseUrl: "http://localhost:8000",
apiKey: "test-key",
});
const result = await client.add({
data: "Test data",
datasetName: "test-dataset",
});
expect(result).toEqual({
datasetId: "test-dataset-id",
datasetName: "test-dataset",
message: "Data added successfully",
});
expect(request).toHaveBeenCalledWith(
"http://localhost:8000/api/v1/add",
expect.objectContaining({
method: "POST",
headers: expect.objectContaining({
Authorization: "Bearer test-key",
"X-Api-Key": "test-key",
}),
}),
);
});
it("should handle errors", async () => {
const mockResponse = {
statusCode: 500,
body: {
text: vi.fn().mockResolvedValue("Internal server error"),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient();
await expect(
client.add({
data: "Test data",
datasetName: "test-dataset",
}),
).rejects.toThrow(
"Cognee add request failed: Cognee add failed with status 500: Internal server error",
);
});
});
describe("cognify", () => {
it("should run cognify successfully", async () => {
const mockResponse = {
statusCode: 200,
body: {
json: vi.fn().mockResolvedValue({
status: "success",
message: "Cognify completed",
}),
text: vi.fn(),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient({
baseUrl: "http://localhost:8000",
});
const result = await client.cognify({
datasetIds: ["dataset-1"],
});
expect(result).toEqual({
status: "success",
message: "Cognify completed",
});
});
});
describe("search", () => {
it("should search successfully", async () => {
const mockResponse = {
statusCode: 200,
body: {
json: vi.fn().mockResolvedValue({
results: [
{
id: "result-1",
text: "Test result",
score: 0.9,
metadata: { path: "test.md" },
},
],
query: "test query",
search_type: "GRAPH_COMPLETION",
}),
text: vi.fn(),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient();
const result = await client.search({
queryText: "test query",
searchType: "GRAPH_COMPLETION",
});
expect(result.results).toHaveLength(1);
expect(result.results[0]).toEqual({
id: "result-1",
text: "Test result",
score: 0.9,
metadata: { path: "test.md" },
});
expect(result.query).toBe("test query");
});
it("should use default search type", async () => {
const mockResponse = {
statusCode: 200,
body: {
json: vi.fn().mockResolvedValue({
results: [],
query: "test",
search_type: "GRAPH_COMPLETION",
}),
text: vi.fn(),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient();
await client.search({ queryText: "test" });
expect(request).toHaveBeenCalledWith(
expect.any(String),
expect.objectContaining({
body: expect.stringContaining('"searchType":"GRAPH_COMPLETION"'),
}),
);
});
});
describe("status", () => {
it("should get status successfully", async () => {
const mockResponse = {
statusCode: 200,
body: {
json: vi.fn().mockResolvedValue({ status: "healthy" }),
text: vi.fn(),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient();
const result = await client.status();
expect(result).toEqual({
status: "healthy",
});
});
});
describe("healthCheck", () => {
it("should return true when status is successful", async () => {
const mockResponse = {
statusCode: 200,
body: {
json: vi.fn().mockResolvedValue({ status: "healthy" }),
text: vi.fn(),
},
};
vi.mocked(request).mockResolvedValue(mockResponse as any);
const client = new CogneeClient();
const result = await client.healthCheck();
expect(result).toBe(true);
});
it("should return false when status fails", async () => {
vi.mocked(request).mockRejectedValue(new Error("Connection failed"));
const client = new CogneeClient();
const result = await client.healthCheck();
expect(result).toBe(false);
});
});
});

View File

@ -1,27 +1,34 @@
import { describe, expect, it, vi, beforeEach } from "vitest";
import { describe, expect, it, vi } from "vitest";
import { CogneeMemoryProvider } from "./cognee-provider.js";
import type { ClawdbotConfig } from "../config/config.js";
const searchMock = vi.fn();
vi.mock("./cognee-client.js", () => {
class CogneeClient {
healthCheck = vi.fn().mockResolvedValue(true);
add = vi.fn().mockResolvedValue({
datasetId: "test-dataset-id",
datasetName: "test-dataset",
message: "Success",
});
cognify = vi.fn().mockResolvedValue({
status: "success",
message: "Cognify completed",
});
search = vi.fn().mockResolvedValue({
search = searchMock;
}
return { CogneeClient };
});
describe("CogneeMemoryProvider", () => {
it("maps search results into memory snippets", async () => {
const mockConfig: ClawdbotConfig = {
agents: {
defaults: {
workspace: "/tmp/test-workspace",
},
},
};
searchMock.mockResolvedValue({
results: [
{
id: "result-1",
text: "Test result text",
text: "A".repeat(800),
score: 0.85,
metadata: {
path: "test.md",
path: "memory/test.md",
source: "memory",
},
},
@ -29,124 +36,51 @@ vi.mock("./cognee-client.js", () => {
query: "test query",
searchType: "GRAPH_COMPLETION",
});
status = vi.fn().mockResolvedValue({
status: "healthy",
version: "1.0.0",
datasets: [
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
const results = await provider.search("test query");
expect(results).toHaveLength(1);
expect(results[0]).toMatchObject({
path: "memory/test.md",
source: "memory",
score: 0.85,
});
expect(results[0].snippet.length).toBeGreaterThan(700);
expect(results[0].snippet.endsWith("...")).toBe(true);
});
it("defaults missing metadata to safe values", async () => {
searchMock.mockResolvedValue({
results: [
{
id: "test-dataset-id",
name: "clawdbot",
documentCount: 5,
id: "result-2",
text: "Short result",
score: 0.2,
},
],
query: "missing metadata",
searchType: "GRAPH_COMPLETION",
});
}
return { CogneeClient };
});
vi.mock("./internal.js", () => ({
listMemoryFiles: vi.fn().mockResolvedValue([]),
buildFileEntry: vi.fn(),
hashText: vi.fn().mockReturnValue("test-hash"),
}));
vi.mock("node:fs/promises", () => ({
default: {
readdir: vi.fn().mockResolvedValue([]),
readFile: vi.fn().mockResolvedValue("Test file content"),
stat: vi.fn().mockResolvedValue({ mtimeMs: 1234567890, size: 100 }),
},
}));
describe("CogneeMemoryProvider", () => {
const mockConfig: ClawdbotConfig = {
agents: {
defaults: {
workspace: "/tmp/test-workspace",
const mockConfig: ClawdbotConfig = {
agents: {
defaults: {
workspace: "/tmp/test-workspace",
},
},
},
};
};
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
beforeEach(() => {
vi.clearAllMocks();
});
const results = await provider.search("missing metadata");
describe("constructor", () => {
it("should initialize with default configuration", () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
expect(provider).toBeDefined();
});
it("should initialize with custom configuration", () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"], {
baseUrl: "http://custom:8000",
apiKey: "custom-key",
datasetName: "custom-dataset",
searchType: "chunks",
maxResults: 10,
});
expect(provider).toBeDefined();
});
});
describe("healthCheck", () => {
it("should perform health check", async () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
const healthy = await provider.healthCheck();
expect(healthy).toBe(true);
});
});
describe("search", () => {
it("should search and transform results", async () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
const results = await provider.search("test query");
expect(results).toHaveLength(1);
expect(results[0]).toMatchObject({
path: "test.md",
source: "memory",
score: 0.85,
snippet: "Test result text",
});
});
it("should respect maxResults setting", async () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"], {
maxResults: 5,
});
const results = await provider.search("test query");
expect(results.length).toBeLessThanOrEqual(5);
});
});
describe("cognify", () => {
it("should run cognify", async () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
await expect(provider.cognify()).resolves.not.toThrow();
});
});
describe("getStatus", () => {
it("should return status information", async () => {
const provider = new CogneeMemoryProvider(mockConfig, "test-agent", ["memory"]);
const status = await provider.getStatus();
expect(status).toMatchObject({
connected: true,
datasetName: "clawdbot",
syncedFileCount: 0,
});
expect(results).toHaveLength(1);
expect(results[0]).toMatchObject({
path: "unknown",
source: "memory",
score: 0.2,
snippet: "Short result",
});
});
});