Run Kilo Code against GLM, Kimi, DeepSeek, Qwen or MiniMax

Kilo Code connects to OpenAI-compatible providers. Point it at A2Agent with a base URL, a key and a model id, and the same key also serves the Anthropic and Gemini protocol faces for whatever else you run.

Setup

1
Create an API key
Sign up and create a key. New accounts start with trial credit, so you can finish the setup without topping up.
2
Configure an OpenAI Compatible provider
Kilo Code asks for the endpoint, the credential and the model to call. These are the three values.
```
Base URL:  https://a2agent.me/v1
API Key:   sk-your-a2agent-key
Model ID:  deepseek-v4-pro
```

Confirm the endpoint answers

Checking with curl first tells you whether a later failure is the extension or the gateway.

curl https://a2agent.me/v1/chat/completions \
  -H "content-type: application/json" \
  -H "authorization: Bearer sk-your-a2agent-key" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [{"role": "user", "content": "Reply with OK"}]
  }'

What differs from the official endpoint

The endpoint is the same one every other client uses

A2Agent serves one gateway with three protocol faces. Whichever editor you configure, the requests land on the same routing, the same billing and the same catalogue — there is no per-client endpoint to get wrong.

Model ids are case-sensitive

Matching is exact, with no case folding. Take the id from the models page rather than retyping it from documentation elsewhere.

Context windows differ by family, not by tier

DeepSeek runs 1M across both its tiers, GLM-5 and GLM-5.1 sit at 200K while GLM-5.2 and later reach 1M, Kimi is 256K except K3, and every Qwen model runs 1M. If a session keeps truncating, change family rather than trimming the prompt.

An unassigned key fails before it reaches an upstream

Keys must belong to a group. A key without one is rejected up front with a structured error, which is worth checking first if the extension reports something auth-shaped right after setup.

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

Does Kilo Code need a different endpoint from Cline or Roo Code?

No. All three speak the OpenAI Chat Completions API and use https://a2agent.me/v1.

Which model has the longest context?

DeepSeek, every Qwen model, MiniMax M3 and M2.5, Kimi K3 and GLM-5.2 and later all run 1M. Kimi K2.x is 256K, and GLM-5 and GLM-5.1 are 200K.

How do I see what a session cost?

The dashboard reports usage and balance per key, and /v1/usage answers the same over the API.

Is a card required to try it?

No. Sign-up grants trial credit.

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

Get an API key Browse models See pricing