Real-time event infrastructure for AI agent systems.

Agents emit events. Those events land durably, fan out to subscribers in real time and to outbound webhooks, and stay searchable and traceable afterwards.

Point an existing setup at axonpush rather than rewriting your instrumentation. Three ingest dialects are supported:

• the native REST API
• OTLP/HTTP, at /v1/traces and /v1/logs
• Sentry SDK envelopes, pointed at an axonpush DSN

from axonpush import AxonPush

with AxonPush() as client:
    client.events.publish(
        identifier="research.step",
        channel_id=CHANNEL_ID,
        payload={"query": "who cited this paper", "tokens": 412},
    )

Subscribe to the same channel over MQTT and the event arrives as it is written.

All three live in sdks and generate from one OpenAPI contract.

The generated client is the floor, not the product. On top of it sit resource classes, a realtime client, and a shared error tree. Integrations cover LangChain, LangGraph, OpenAI Agents, Anthropic, CrewAI, Mastra, LlamaIndex, Vercel AI, Semantic Kernel, OpenTelemetry, and the usual logging libraries.

Cross-language drift is checked rather than hoped for. CI compares the resource surfaces across languages, so a method that exists in one SDK and not another fails the build.

npx @axonpush/wizard

This installs skills into your coding agent. The agent then reads your project, picks the right integration, provisions credentials, and writes .env. It finishes by publishing a test event and reading it back, so you find out immediately if the path is wrong.

Works in Claude Code, Cursor, Codex, and around fifty other agents.

• axonpush.xyz
• Dashboard
• Documentation