Simple, affordable observability for your AI agents. Drop in the SDK. See your traces. Get alerts when agents fail. No self-hosting, no enterprise contracts.
Free forever up to 1,000 traces/month
Works with your agent stack
Full waterfall view of your agent's steps — LLM calls, tool uses, sub-agent invocations. Know exactly what happened and when.
Email alerts the moment a trace errors or times out. With rate limiting to prevent alert fatigue. Know before your users do.
Open-source TypeScript and Python SDKs. Drop into any agent — Claude, GPT-4, LangChain, custom. Start seeing traces in under 5 minutes.
From zero to first trace in under 5 minutes.
TypeScript or Python — one install in your agent project.
npm install @keylightdigital/nexus
pip install keylightdigital-nexus
Wrap your agent run in a trace. Spans auto-capture timing.
Every run shows up at nexus.keylightdigital.dev — waterfall view, error rates, timing.
Works with Claude, GPT-4, LangChain, or raw fetch calls — in TypeScript or Python.
import { NexusClient } from '@keylightdigital/nexus'
const nexus = new NexusClient({
apiKey: 'nxs_your_key_here',
agentId: 'my-assistant'
})
// Start a trace when your agent begins
const trace = await nexus.startTrace({ name: 'process-request' })
await trace.addSpan({ name: 'call-llm', output: llmResponse })
await trace.end({ status: 'success' })from nexus_agent import NexusClient
nexus = NexusClient(
api_key='nxs_your_key_here',
agent_id='my-assistant'
)
# Start a trace when your agent begins
trace = nexus.start_trace(name='process-request')
trace.add_span(name='call-llm', output=llm_response)
trace.end(status='success')From indie devs and small teams shipping AI agents in production.
“Dropped Nexus into my CrewAI pipeline in 10 minutes. Caught a silent tool failure that had been corrupting 8% of runs for two weeks. That alone paid for a year of Pro.”
Alex K.
Freelance AI engineer
“I tried self-hosting Langfuse for a weekend and gave up. Nexus was running in my LangChain agent before lunch on Monday. The waterfall view is exactly what I needed to find the slow span.”
Sam P.
Solo founder, AI startup
“We run 5 agents in production and the per-agent health cards are perfect. Saw our summarisation agent spike to 12% errors on Friday night — fixed it before any user complained.”
Maya R.
ML engineer, 4-person team
No enterprise tiers. No seats. No surprises.
Start with the free tier. Upgrade when you need more traces, more agents, or email alerts.
| Feature | Free | Pro |
|---|---|---|
| Trace viewer | ✓ | ✓ |
| Span waterfall | ✓ | ✓ |
| Dashboard & metrics | ✓ | ✓ |
| Data retention | 30 days | 90 days |
| Agents | 1 | Unlimited |
| Monthly traces | 1,000 | 50,000 |
| Email alerts on failure | — | ✓ |
| Team access | — | Up to 5 |
| Support | Community | Priority |
Nexus was designed and built by Ralph, an autonomous AI agent employed by Keylight Digital LLC. Ralph needed a way to monitor his own agent runs — so he built the tool he wished existed.
Every feature in Nexus was motivated by a real problem Ralph encountered managing AI workflows: knowing when a trace failed, understanding why an agent stalled, tracking performance over time. If you're building with AI agents, you have the same problems. Nexus is for you.
pip install keylightdigital-nexus. It mirrors the TypeScript SDK's API surface with NexusClient, start_trace(), add_span(), and end(). Works with any Python agent framework — LangChain, AutoGen, CrewAI, or custom agents. See the docs for the full Python quickstart.
429 with a clear error message and a link to upgrade. Your existing traces are never deleted — they stay visible in your dashboard until the 30-day retention window. Upgrading to Pro ($9/mo) instantly raises the limit to 50,000 traces.
deploy.sh script. The hosted version at nexus.keylightdigital.dev is the easiest option — but self-hosting is fully supported.
Start free. No credit card. Under 5 minutes to first trace.
Also from Keylight Digital
Beam — Privacy-first web analytics. Cookie-free, GDPR-compliant. ↗