The Plausible of AI agents. Simple, affordable, open-source control plane for your AI agents. Drop in the SDK. See your traces. Get alerts when they fail. No self-hosting, no enterprise contracts.
Built by an AI agent (Ralph), for AI agents. Keylight Digital built this because we needed it to manage our own agent infrastructure.
nexus.keylightdigital.dev · Free forever · Pro at $9/mo
npm install @keylightdigital/nexusimport { NexusClient } from '@keylightdigital/nexus'
const nexus = new NexusClient({ apiKey: 'nxs_...', agentId: 'my-assistant' })
const trace = await nexus.startTrace({ name: 'process-invoice' })
await trace.addSpan({ name: 'extract-data', output: { invoice_id: '123' } })
await trace.addSpan({ name: 'call-gpt-4o', input: { prompt }, output: { result } })
await trace.end({ status: 'success' })pip install keylightdigital-nexusfrom nexus_agent import NexusClient
nexus = NexusClient(api_key='nxs_...', agent_id='my-assistant')
trace = nexus.start_trace(name='process-invoice')
trace.add_span(name='extract-data', output={'invoice_id': '123'})
trace.add_span(name='call-gpt-4o', input={'prompt': prompt}, output={'result': result})
trace.end(status='success')Get your API key at nexus.keylightdigital.dev. That's it.
Enterprise observability tools cost $39–$100/mo and are built for teams of 50. If you're an indie developer building an AI agent, you need:
- A trace viewer that shows what your agent actually did
- Email alerts when it fails at 3am
- A usage dashboard so you're not flying blind
- Something that costs less than your coffee budget
Nexus is that tool. Built on Cloudflare — near-zero latency globally, no servers to maintain.
| Free | Pro | |
|---|---|---|
| Price | $0 | $9/mo |
| Traces/month | 1,000 | 50,000 |
| Agents | 1 | Unlimited |
| Retention | 30 days | 90 days |
| Email alerts | — | ✓ |
| Team access | — | Up to 5 |
- Trace viewer — Waterfall timeline of every span: inputs, outputs, errors, timing
- Agent health dashboard — Error rate, avg latency, 7-day volume sparkline with live auto-refresh
- Trace filtering & search — Filter by status, agent, date range; search by trace name or metadata
- Email alerts — Notified when an agent trace ends with
errorortimeout(Pro) - OpenTelemetry compatible — Point any OTEL exporter at
/v1/traces— no code changes needed - Public trace API —
GET /api/v1/tracesfor custom dashboards, Slack bots, CI/CD integrations - Open-source SDKs — TypeScript and Python, non-throwing, minimal dependencies
- Multi-agent — Track as many agents as you want (Pro), or start with 1 (Free)
- Privacy-aware — Cloudflare-native, data stays in your D1 database
Nexus is built entirely on the Cloudflare stack:
- Runtime: Cloudflare Workers (Hono framework)
- Database: D1 (SQLite at the edge)
- Cache: KV (sessions, rate limits, trace counts)
- Email: Resend API
Zero cold starts. Global edge. ~$0 infrastructure cost at indie scale.
Ready-to-run example projects are in the scobb/nexus-examples repo:
| Example | Description |
|---|---|
| Node.js / TypeScript | TypeScript agent with trace+span creation, error handling |
| Python | Python agent with sync and async patterns |
| LangChain | LangChain agent with automatic LLM call and tool tracing |
git clone https://github.com/scobb/nexus-examples
cd nexus-examples/examples/nodejs
npm install && NEXUS_API_KEY=nxs_... npm startThe @keylightdigital/nexus SDK is open-source and available on npm. See sdk/README.md for full docs.
This entire product — PRD, architecture, implementation, SDK, and this README — was written by Ralph, an AI agent built by Keylight Digital LLC.
Ralph built Nexus because Ralph needed it. Our AI agents were running blind — no visibility into what they were doing, why they were failing, or how long they were taking. Enterprise tools were $100/mo. Self-hosting Langfuse required a VPS. There was no Plausible for AI agents.
So Ralph built one. In about 20 stories. On a $100 budget.
See CONTRIBUTING.md for setup instructions, development workflow, and PR guidelines.