Point Roo Code at a cheaper OpenAI-compatible endpoint

Roo Code drives any provider that speaks the OpenAI Chat Completions API. Give it the A2Agent base URL, a key and a model id, and the extension runs against GLM, Kimi, DeepSeek, Qwen or MiniMax instead of a frontier vendor.

Setup

1
Create an API key
Sign up and create a key in the dashboard. Trial credit is granted on sign-up, so the first requests cost nothing.
2
Add an OpenAI Compatible provider
Roo Code keeps provider profiles in its settings. Create one of the OpenAI Compatible kind and fill in these three values.
```
Base URL:  https://a2agent.me/v1
API Key:   sk-your-a2agent-key
Model ID:  glm-5.1
```

Verify outside the editor first

Extensions tend to swallow provider errors. One curl separates a bad key from a bad model id.

curl https://a2agent.me/v1/chat/completions \
  -H "content-type: application/json" \
  -H "authorization: Bearer sk-your-a2agent-key" \
  -d '{
    "model": "glm-5.1",
    "messages": [{"role": "user", "content": "Reply with OK"}]
  }'

What differs from the official endpoint

Model ids are matched exactly

The gateway looks the requested model up in the account mapping without folding case, so glm-5.1 and GLM-5.1 are not interchangeable. Copy ids from the models page. This is the single most common cause of a provider error that looks like an auth problem.

Agent loops burn output tokens, so pick the tier accordingly

Roo Code runs multi-step loops that re-read their own output. Output rates matter more than input rates here: within a family the cheaper tier often costs a third of the flagship for the same context window, which changes the economics of a long session more than the model choice does.

One key covers every family

Switching the model id is enough to switch upstream. The gateway resolves the group per request, so you do not keep a separate provider profile per vendor.

Billing is per token against a balance

No monthly fee, no minimum spend, and credit does not expire. The rates on the models page come from the same billing tables the gateway charges against.

Models worth starting with

Any model in the catalogue works through this client. These are the ones with a full page covering pricing, context window and setup.

Questions

Which provider type do I pick in Roo Code?

OpenAI Compatible, with https://a2agent.me/v1 as the base URL.

Why does my model id fail?

Almost always the case is wrong. Ids are matched exactly, so copy them from the models page.

Which model should I start with?

A coding-tagged model with a long window, such as GLM-5.3 at 1M. For agent loops that accumulate tool output, MiniMax M3 gives a 1M window at a lower rate.

Can I share the key with other editors?

Yes. The same key serves the Anthropic, OpenAI and Gemini faces, so Claude Code, Codex CLI, Cline and Roo Code can all use one credential.

Is there a free tier?

New accounts get trial credit and no card is required to start.

Trial credit on sign-up, billed per token, no minimum spend.

Get an API key Browse models See pricing