Start here
Get your first agent trace into axonpush. Pick the way in that your stack already speaks, an SDK import, an OpenTelemetry collector, or a Sentry DSN you already have.
axonpush is an observability platform for AI agents and the backends behind them. Agent events, backend logs and OpenTelemetry spans land on one timeline, joined by a single trace ID, so a whole request reads as one trace.
Your first trace
Install the SDK
pip install axonpushPoint it at your project
Create a key in the dashboard, then construct the client with it. The SDK wraps the client you already build, there is no separate transport to manage.
from axonpush import AxonPush
client = AxonPush(api_key="ak_...")Watch it land
Run your agent, then open the Observe pillar in the dashboard. The trace is there, with every model call, tool call and log line underneath it on one waterfall.
Already on OpenTelemetry?
Then there is no SDK to install. Point your existing collector at
api.axonpush.xyz over OTLP/HTTP and your spans arrive live, two
environment variables and no code at all. See OTLP ingest.
Pick your path
Let your coding agent do it
npx skills add axonpush/skills, and Claude Code, Cursor or Codex reads your stack, picks the right integration and writes the wiring.
Connect over MCP
Point Claude Code, Cursor or any MCP client at the axonpush MCP server. It provisions your project and reads traces back, no key to paste.
Python SDK
Non-blocking handlers for stdlib logging, Loguru, structlog and OpenTelemetry. FastAPI, Django, Flask and AWS Lambda.
TypeScript SDK
Integrations for Pino, Winston, console capture and OpenTelemetry, across Node, Bun and serverless runtimes.
.NET SDK
Semantic Kernel telemetry in one call, an OpenTelemetry span exporter, or the raw events client.
API reference
Every endpoint, generated from the contract the server actually serves.
What you get once it is wired
One timeline, both halves
The model call, the tool call it triggered, the HTTP request underneath and the query that timed out, on one waterfall, joined by one trace ID.
Alerts and moderation, built in
Fire an alert when a trace crosses a threshold, and enforce moderation rules over the content flowing through your agents.
Nothing proprietary
OpenTelemetry in, OpenTelemetry out. An existing Sentry DSN works unchanged. Leaving is a config line too.
Your cloud, if it has to be
Self-host runs the whole platform inside your own AWS account, with traces that never leave it.