The AI memory layer. Three ways to connect.

Step 1 - Get an API key

Dashboard > Settings > API Keys > New Key. Keys are prefixed sx_live_. Store it in your environment because it is shown once.

# Verify your key
curl https://synchronex.ai/api/v1/memory/identity \
  -H "Authorization: Bearer sx_live_YOUR_KEY"

Step 2 - Install the SDK

npm install @synchronex/sdk

Step 3 - Read your Identity Kernel

import { Synchronex } from '@synchronex/sdk'

const sx = new Synchronex({ apiKey: process.env.SYNCHRONEX_API_KEY })
const kernel = await sx.memory.identity.get()
console.log(kernel.injection_block)

• Keys are shown once at creation. Store immediately.
• Keys can be rotated without downtime. The old key stays valid for 60 seconds after rotation.
• REST API keys use the sx_live_* prefix.
• OAuth tokens expire after 1 hour. Refresh tokens rotate on use.

curl -X DELETE https://synchronex.ai/api/v1/byok/KEY_ID \
  -H "Authorization: Bearer sx_live_YOUR_KEY"

A standard MCP server exposes tools derived from the canonical registry. Connect once; every AI client on that machine can access the scope you grant.

// .claude/mcp.json or ~/.claude/settings.json
{
  "synchronex": {
    "command": "npx",
    "args": [
      "-y",
      "@synchronex/memory-mcp"
    ],
    "env": {
      "SYNCHRONEX_API_KEY": "sx_live_YOUR_KEY"
    }
  }
}

MCP Tool Reference

Connection Registry

Connection names, categories, auth type, and live status sources come from the canonical registry used by Settings, the guide, and consent-adjacent labels.

Example agent conversation

User: "What database did I decide on?"
Agent calls: search_decisions(q: "database decision")
Agent: "You chose Supabase (Postgres + pgvector) for native RLS and auth.uid() integration. Decision logged 2026-04-21."

Clients that support MCP auto-discovery can locate the server at GET https://synchronex.ai/.well-known/mcp.

Agent Bridge lets Synchronex dispatch work to a locally running AI CLI on your machine. The bridge streams execution events back over HTTP polling. Use it when you need cloud-triggered local AI execution, such as automated PR reviews or scheduled kernel refreshes that must run in your local dev context.

New here? The bridge setup guide walks through installing and running the daemon (npx, one-line script, or standalone binary) and serving multiple accounts from one machine. The REST reference below is for triggering bridge work programmatically.

# 1. Create a bridge request
curl -X POST https://synchronex.ai/api/agent-bridge/requests \
  -H "Authorization: Bearer sx_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Summarize open decisions for my-app","product_name":"my-app"}'

# 2. Poll for events (replace REQUEST_ID)
curl "https://synchronex.ai/api/agent-bridge/requests/REQUEST_ID/events?after=0" \
  -H "Authorization: Bearer sx_live_YOUR_KEY"

SDK methods for Agent Bridge, such as sx.bridge.run(), are planned. The current path is direct REST as shown above.

Base URL: https://synchronex.ai/api/v1. Auth: Authorization: Bearer sx_live_YOUR_KEY.

GET /v1/memory/identity

{
  "id": "uuid",
  "user_id": "uuid",
  "injection_block": "# Identity\nName: ...\n...",
  "confidence": 0.94,
  "token_count": 312,
  "updated_at": "2026-05-01T00:00:00Z"
}

PUT /v1/memory/identity

curl -X PUT https://synchronex.ai/api/v1/memory/identity \
  -H "Authorization: Bearer sx_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name":"Kevin Desler","role":"Solo founder","timezone":"Pacific/Honolulu"}'

{
  "decision": "Chose Supabase over PlanetScale for multi-tenant DB",
  "reasoning": "Native RLS, auth.uid() integration, pgvector bundled",
  "product_name": "synchronex",
  "affects_products": []
}

curl "https://synchronex.ai/api/v1/memory/decisions/search?q=database+choice&limit=5" \
  -H "Authorization: Bearer sx_live_YOUR_KEY"

GET /v1/kernel/export

curl "https://synchronex.ai/api/v1/kernel/export?format=claude-md&budget=400" \
  -H "Authorization: Bearer sx_live_YOUR_KEY"

curl "https://synchronex.ai/api/v1/kernel/export?format=gemini-md" \
  -H "Authorization: Bearer sx_live_YOUR_KEY"

Bring Your Own Key stores encrypted credentials for AI providers and third-party APIs. Requires Pro tier or above. Keys are stored AES-256-GCM encrypted and never returned in plaintext after submission.

{
  "service": "anthropic",
  "label": "Primary Claude API key",
  "key": "sk-ant-..."
}

Supported providers: anthropic openai google openrouter resend linear sentry stripe custom

Requires Pro tier or above.

Controls the active workspace context.

{
  "active_workspace_id": "workspace-uuid"
}

Connect Synchronex to hosted AI tools via OAuth with no API key in the AI client. Supported clients include Claude.ai, ChatGPT Apps, and Gemini. Supported flows are Authorization Code + PKCE and Dynamic Client Registration.

Available scopes

Access tokens last 1 hour. Refresh tokens last 30 days, rotate on use, and have a 60 second grace window.

Connecting Claude.ai

In Claude.ai > Integrations > Add Integration, paste https://synchronex.ai/.well-known/oauth-authorization-server.

Connecting ChatGPT

Push events from external tools into Synchronex for the cross-project command center.

Path 1 - Native webhook receivers

Path 2 - MCP log_project_event tool

{
  "name": "log_project_event",
  "arguments": {
    "project_id": "uuid",
    "event_type": "deployment.completed",
    "title": "Deployed v2.3.1 to production",
    "severity": "info",
    "body": {
      "sha": "abc123",
      "environment": "production"
    }
  }
}

Path 3 - Generic REST endpoint

Send HMAC-signed events to /api/v1/ingest/:projectId when a native webhook receiver is not available.

curl -X POST https://synchronex.ai/api/v1/ingest/PROJECT_ID \
  -H "X-Synchronex-Signature: sha256=COMPUTED_HMAC" \
  -H "X-Synchronex-Timestamp: 1746057600" \
  -H "Content-Type: application/json" \
  -d '{"event_type":"build.failed","title":"CI build failed on main","severity":"error","body":{"branch":"main"},"external_id":"ci-run-9876"}'

Signed payload format is {timestamp}.{rawBody}. Timestamp tolerance is +/- 5 minutes. Pass external_id for idempotency; duplicate delivery returns 200 with no duplicate row.

All limits are per-user, not per-IP, for authenticated calls.

{
  "error": "rate_limit_exceeded",
  "retry_after": 30,
  "limit": 60,
  "remaining": 0,
  "reset_at": "2026-05-03T12:01:00.000Z"
}

SDK retry behavior: 3 attempts, exponential backoff starting at 1 second, respects the Retry-After header.

npm install @synchronex/sdk

import { Synchronex } from '@synchronex/sdk'

const sx = new Synchronex({ apiKey: process.env.SYNCHRONEX_API_KEY })

const kernel = await sx.memory.identity.get()

await sx.memory.decisions.log({
  decision: 'Chose Postgres over MongoDB',
  reasoning: 'Native JSON + pgvector, no extra vendor',
  product_name: 'my-app',
})

const claudeMd = await sx.kernel.export({ format: 'claude-md', budget: 400 })
const results = await sx.memory.decisions.search({ q: 'database decisions', limit: 5 })
console.log(kernel, claudeMd, results)

import { SynchronexError } from '@synchronex/sdk'

try {
  await sx.memory.identity.update({ name: 'Kevin' })
} catch (err) {
  if (err instanceof SynchronexError) {
    console.error(err.code, err.message, err.status)
  }
}

Machine-readable OpenAPI 3.1 is available as JSON and YAML.

• JSON: https://synchronex.ai/api/v1/openapi.json
• YAML: https://synchronex.ai/api/v1/openapi.yaml or https://synchronex.ai/openapi.yaml

npx openapi-typescript https://synchronex.ai/api/v1/openapi.json -o synchronex.d.ts