Overview
Use Tensor Machine models inside the tools you already run — coding agents, CLIs, SDKs and IDEs — by pointing them at your workspace endpoint.
Most AI tools let you choose where their requests go. Because a Tensor Machine key speaks both the OpenAI Chat Completions format and the Anthropic Messages format over one endpoint, almost any tool that talks to either can be pointed at your workspace instead — usually with a base URL and an API key, no code changes.
That also means the tool's traffic follows your workspace rules: your model allow-list, your spend limits, your audit trail, and India-resident inference.
Which format does your tool speak?
Pick the endpoint that matches the API your tool was built against.
| Your tool speaks | Point it at | Auth header it will send |
|---|---|---|
| OpenAI Chat Completions | https://edge.tensormachine.ai/<org>/v1 | Authorization: Bearer sk-tm-… |
| Anthropic Messages | https://edge.tensormachine.ai/<org>/anthropic | x-api-key or Authorization: Bearer |
Replace <org> with your workspace slug — the exact base URL is on the API Keys page in
the console, next to a copy button.
Either header works on the Anthropic endpoint, which matters because tools disagree about
which one to send: the Anthropic SDK uses x-api-key, while Claude Code sends
Authorization: Bearer. You don't have to care which — the same sk-tm-… key authenticates
both ways.
Guides
- OpenCode — the open-source terminal coding agent. Uses the
OpenAI-compatible endpoint, configured in
opencode.json. - Claude CLI — Anthropic's
claudecommand-line tool. Uses the Anthropic-compatible endpoint, configured with two environment variables.
Before you start
Every guide assumes you have:
- An API key — mint one under API Keys in the console and copy the
sk-tm-…secret (it's shown once). See API keys & endpoints. - A funded workspace — a key can't make requests at a ₹0 balance (Billing → Add funds).
- At least one model allowed for your team — check My Team → Model access. The model
IDs you'll paste into these configs are the
tm/…aliases listed there.
A note on agentic tools
Coding agents are demanding clients: they send long system prompts, and they depend on tool calling to read and edit files. A model that answers chat well can still be a poor agent if it handles tool calls badly. If an agent connects but loops, ignores instructions, or never edits a file, try a different model from your allow-list before assuming the integration is misconfigured — Models notes which models are suited to this.
Something else you want connected?
These guides cover the tools we've been asked about most. If you're wiring up something not listed here and get stuck — or you get it working and think it deserves a page — tell us through the console and we'll add it.