Jents Academy

From sign-up to seeing every dollar your AI spends: import the bill, capture Claude Code and Codex over OpenTelemetry, track seats, set budgets.

The master reference for every way data reaches Jents — admin-key bill imports, telemetry, your own gateway — source by source, in one table, including what no source can tell you today. Plus the one that matters most: the gap between who uses a tool and who you pay for is the money you're wasting.

Every tool, your own agents, Claude Code, Cowork, Codex, ChatGPT and Claude seats, how each connects and what you can measure, in one table — and, plainly, the ones you can't connect today. Plus the two rules for ROI.

Backend service, chatbot, RAG, serverless, batch job, embeddings, or a script, point your AI client at Jents (Python & Node).

Override the base URL on the tool's built-in OpenAI connection so every step routes through Jents.

Set the base URL on the model once, every agent, tool, and retry routes through Jents.

Point the client's base URL + key at Jents; threads and runs flow through the gateway.

Codex exports OpenTelemetry, so one block in ~/.codex/config.toml gives you per-person usage on any Codex, including a ChatGPT subscription. Nothing changes on the call path, and the three settings Codex fails silently on.

One admin setting turns on capture for the whole org — nothing for teammates to paste. Cowork reports its own exact cost, so nothing is estimated, and every call is already attributed to a person.

Connect the account where your AI bills — OpenAI or Anthropic — and Jents imports an agent per API key, your people, and your real cost automatically. What the bill can't tell you, and why the per-key dollars are apportioned while the invoice total stays exact.

Already run a self-hosted LiteLLM? Connect it — Jents reads its exact cost + usage, and can control it too. No re-routing, nothing to register first; the enterprise data-residency answer.

Two ways to get your AI into Jents: an account import is your bill (broad, zero-setup, daily); a gateway is a checkpoint (deep, live, controllable). A side-by-side of what each gives you, and why it's a staircase, not a choice.

Claude Code, Claude Cowork and OpenAI Codex report per-person usage over telemetry; ChatGPT Team and Claude seats arrive with the same Admin key that imports your bill. And, plainly: Cursor and GitHub Copilot show nothing at all today.