Provider integration

Use DeepSeek in Codex, Claude Code and any coding agent

DeepSeek speaks Chat Completions. Codex speaks Responses and Claude Code speaks Anthropic Messages. AutoJev sits in the middle on localhost and converts both ways, so every agent can run on your DeepSeek key.

Built-in preset: https://api.deepseek.com
Codex → Responses → Chat conversion, streaming and tools included
Claude Code → Messages → Chat conversion
Mix DeepSeek with other models in one route

Download AutoJev Star on GitHub

Providers

+ Providers

OpenRouter

https://openrouter.ai/api

DeepSeek

https://api.deepseek.com

OpenAI

https://api.openai.com/v1

Ollama

http://127.0.0.1:11434

Why DeepSeek needs a translator

DeepSeek is a popular choice for routine coding turns because it is inexpensive, but its main API is OpenAI Chat Completions. Codex now sends OpenAI Responses requests, and Claude Code sends Anthropic Messages. Pointing either agent straight at api.deepseek.com does not work without a translation layer in between.

AutoJev is that layer, running on your own machine. You add DeepSeek once with its Chat Completions API type. When Codex calls the gateway, AutoJev translates the Responses request and the tool history into Chat Completions, calls DeepSeek, and streams the answer back as Responses events. Claude Code gets the same treatment through /v1/messages.

Add DeepSeek and connect an agent

1
Create the DeepSeek provider
Providers → Add provider → DeepSeek. The base URL is preset. Keep the API type on OpenAI Chat Completions and paste your DeepSeek key.
text
Provider type: DeepSeek Provider name: deepseek Base URL: https://api.deepseek.com API type: OpenAI Chat Completions API key: sk-… Test model: deepseek-chat
2
Test it
Click Test. AutoJev sends one minimal request to https://api.deepseek.com/v1/chat/completions and reports success or the upstream error. A reply that contains only reasoning still counts as a pass.
3
Add the models you use
Add each DeepSeek model with the model ID from DeepSeek's documentation (for example `deepseek-chat`), API type Chat Completions, prices per 1M tokens, tier and context window. Agents will see it as `deepseek/deepseek-chat`.
4
Connect Codex or Claude Code
On Agents, select the DeepSeek model or a route containing it and click Connect. AutoJev backs up the agent config, rewrites the base URL and model, and restores the original on disconnect.

How each agent reaches DeepSeek

Agent	Gateway endpoint	Path to DeepSeek
Codex	POST /v1/responses	Responses → Chat Completions, converted back to Responses events
Claude Code	POST /v1/messages	Messages → Chat Completions, converted back to Messages events
Hermes	POST /v1/chat/completions	Native passthrough
OpenCode, Kimi, Grok Build, OpenClaw, OMP	Chat or Messages	Native when the agent uses Chat
Gemini CLI	POST /v1beta/models/…	Gemini bridge → Chat Completions

Try it from the terminal

Any OpenAI-compatible client can call DeepSeek through the gateway. The key is a local placeholder that labels the request in Logs; your real DeepSeek key never leaves AutoJev.

Responses request, answered by DeepSeek

curl http://127.0.0.1:9527/v1/responses \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer autojev-local-terminal' \
  -d '{"model": "deepseek/deepseek-chat", "input": "Explain this regex: ^\\d{4}-\\d{2}$"}'

Use DeepSeek where it fits

A cheap route

Create `autojev/cheap` with DeepSeek plus a fallback model. Load balancing spreads traffic by weight; failover moves to the next candidate on 429 or 5xx before output starts.

Lower cost preference

In an intelligent route set to Lower cost, AutoJev picks the cheapest candidate that meets the request’s capability requirements, using the prices you entered.

See the savings

Set a cost baseline model and Usage estimates how much each DeepSeek request cost compared with that baseline for the same tokens.

Two things conversion will not do

Thinking blocks are dropped when protocols are converted, and hosted tools such as Codex web search are rejected. If a reasoning model needs its private reasoning state passed back with tool results, a full cross-protocol agent loop is not guaranteed. DeepSeek also offers an Anthropic-compatible endpoint (https://api.deepseek.com/anthropic) that you can add with the Anthropic Compatible preset for native Messages.

Six conversion directions

Chat Completions, Responses and Anthropic Messages convert in every direction, including streaming, function tools, images and usage.

One endpoint, three protocols

127.0.0.1:9527/v1

Chat Completions

Responses

Messages

Chat Completions

/v1/chat/completions

—

Responses

/v1/responses

—

Messages

/v1/messages

—

ResponsesDeepSeek · Chat

MessagesOpenAI · Responses

ChatAnthropic · Messages

Frequently asked questions

Can I use DeepSeek in Codex?

Yes. Add DeepSeek in AutoJev with the Chat Completions API type and connect Codex. Codex sends Responses requests to the gateway; AutoJev converts them to Chat Completions for DeepSeek and converts the stream back.

Can I use DeepSeek in Claude Code?

Yes. Claude Code calls /v1/messages on the gateway, which converts to Chat Completions. Alternatively, add DeepSeek’s Anthropic-compatible endpoint as an Anthropic Compatible provider to use Messages natively.

Do tool calls work with DeepSeek through AutoJev?

Function tool definitions, tool calls and results are converted with stable call IDs, in streaming and non-streaming mode. Whether a given model uses tools well is up to the model.

Why does my DeepSeek test pass with an empty answer?

Reasoning models can spend the short test budget on reasoning. AutoJev treats a reasoning-only reply as a pass because the URL, key and model ID were accepted.

Does AutoJev charge for DeepSeek usage?

No. You bring your own DeepSeek key, requests go directly to DeepSeek, and AutoJev adds no markup. AutoJev itself is free.