Proxon SDK for Agents

The Proxon SDK for Agents adds cost and token telemetry to a Node app you control. You install it, launch your app with one flag, and every LLM call the process makes reports its cost and token usage into Proxon, with no changes at the call site. It instruments the Anthropic and OpenAI SDKs in place, so you can slice spend by agent, task, and owner without touching a single request. It is fire-and-forget. It never throws into or slows your request path, and it never sees prompt or response content, only token counts and metadata.

This takes about a minute. You mint an organization credential, install the package and launch with the register flag, and confirm the cost lands in Cost Intelligence. You need to be an organization owner or admin to mint the credential. Anyone can wire up the SDK once they have it.

npm i @proxon/agent-observer

# Where to send agent-cost events, which is Proxon's production ingest host.
export PROXON_OBSERVER_INGEST_URL="https://ingest.proxon.ai"

# Authenticate with the service credential you minted in Step 1.
export PROXON_OBSERVER_SERVICE_CREDENTIAL="prx_svc_<prefix>_<secret>"

# Name the agent this process reports as, since the rollup keys on it.
export PROXON_OBSERVER_AGENT="my-agent"

node --import @proxon/agent-observer/register app.js

[proxon-agent-observer] active — reporting agent-cost to https://ingest.proxon.ai as agent "my-agent"

Attributing spend by agent, task, or customer

The instrumentation cannot know what a given call is for, since that is business context you supply. Wrap a unit of work in withContext, and every event emitted inside it inherits the attribution. The fixed fields, which are agent, task_type, run_type, owner, linear_ticket, and pr_number, become first-class dashboard dimensions. Anything you put in the open-ended attributes bag is discovered automatically as its own breakdown, so a couple of extra keys give you a by-tenant, by-feature, or by-environment view at no extra effort.

import { withContext, flush } from '@proxon/agent-observer';

await withContext(
  {
    // Fixed fields, which become first-class dashboard dimensions.
    agent: 'support-bot',
    task_type: 'triage',
    run_type: 'orchestrator',
    owner: 'octocat',
    linear_ticket: 'PROJ-42',
    pr_number: 128,
    // Open-ended custom dimensions. Attach whatever your business cares about,
    // and each key becomes its own breakdown with no schema change needed.
    attributes: { tenant: 'acme', feature: 'search', environment: 'prod' },
  },
  async () => {
    await anthropic.messages.create({ /* ... */ }); // auto-attributed
  }
);

await flush(); // ship the final batch on shutdown or after a one-shot run

How cost is computed

Providers return token counts, not dollars. Proxon recomputes cost on its own servers from a maintained price book, so you never ship pricing in your agent. Content, meaning prompts and responses, is never sent.