Local AI Guide

See the speed before choosing hardware

Executable programs and extensions

Connect Strata to coding agents: OpenAI, Anthropic, and Codex APIs

Strata exposes OpenAI- and Anthropic-compatible APIs on localhost so coding tools can send requests to a local model.

Strata supports OpenAI Chat Completions and Anthropic Messages, plus the OpenAI Responses API1 starting in Strata 0.1.39. By default, the server accepts requests at 127.0.0.1:8080 and returns Qwen3.8-Flash-Next answers. A coding agent on the same PC uses the matching base URL; Codex CLI2 can connect through /v1/responses. Check /health and /v1/models, then send a short real POST request. Keep testing on loopback; do not change the host to 0.0.0.0 to expose it to a network.

Contents
Requirements and key details
  • The default address is http://127.0.0.1:8080, with separate Chat Completions, Anthropic Messages, and Responses API paths.
  • The Responses API is supported from Strata 0.1.39, and Codex CLI uses this endpoint.
  • First read local health and model endpoints, then verify a real short request and response.

What Strata’s local API accepts and returns

Strata runs Qwen3.8-Flash-Next on a local PC and returns API requests as answers. A coding agent sends conversation messages and tool definitions; the response contains model text and, where supported, tool calls3. Strata does not install the coding agent itself: configure the model-server address in a client you already use.

For a client on the same computer, use http://127.0.0.1:8080. Chat Completions uses /v1/chat/completions, Anthropic Messages uses /v1/messages, and Responses API uses /v1/responses. The default loopback address is not reachable from another computer, so testing does not require exposing the server to a network.

First start the server using the project’s installation guide, then choose an API format supported by your coding tool. Responses API support begins with Strata 0.1.39; on earlier versions, use a client that supports Chat Completions.

A black tower PC and monitor with RAM modules beside them
Strata returns local model answers to a coding tool through its server.

Check health and model listing on localhost

After the server window reports that it is ready, run curl http://127.0.0.1:8080/health in a terminal. If it returns a status response, run curl http://127.0.0.1:8080/v1/models to see the loaded model ID. If you configured an API key, include an Authorization: Bearer <key> header with both requests.

If the model list is empty or the connection fails, check that Strata is still running and that setup did not assign a different port. If another program already uses port 8080, use the port configured for Strata in each request. If no model is loaded, first send a short prompt4 through browser chat to confirm setup has completed.

A graphics card, RAM modules, and an NVMe SSD on a desk
First verify that local health and model-list endpoints respond.

Send a real Chat Completions request

Once the model list responds, send a short prompt with the request below. Put the ID returned by /v1/models in model and the conversation in messages. A successful response contains the model answer in choices[0].message.content.

Chat Completions request
curl http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"strata","messages":[{"role":"user","content":"이 서버가 답하면 Strata 연결이 된 거예요."}],"max_tokens":128}'
Replace `model` with the ID returned by `/v1/models`.

Match the API address and model ID to your client

For an OpenAI-compatible client, set the base URL to http://127.0.0.1:8080/v1. Strata accepts the requested model name, but use an ID from the model list if the client validates names first. Claude Code uses the Anthropic /v1/messages path, so set ANTHROPIC_BASE_URL=http://127.0.0.1:8080. It may also require an ANTHROPIC_MODEL name recognized by the client; that is a client-side setting for routing requests to the local endpoint.

Codex CLI can use the Responses API (/v1/responses) starting with Strata 0.1.39. This API is stateless, so each request must include the conversation input; server-side storage through previous_response_id is not supported. If a client lacks Responses API support or does not interoperate, use another client that supports Chat Completions.

An API response confirms connectivity. For a coding agent to read or edit files or run commands, the client must provide those tools and the user must authorize them. Start with read-only actions or a small test repository, then inspect tool calls and results.

Two graphics cards mounted in an open PC workbench
Choose the API route that matches the client’s supported format.

Terminology notes

  1. API — A defined interface that lets other code call a program’s functions.

    Back to the text

  2. CLI — Short for Command-Line Interface: operating a program by entering commands in a terminal.

    Back to the text

  3. Tool call — A structured request from a model for an external function such as reading a file, searching, or running a command. The agent runtime and its permission settings decide whether the request is actually executed.

    Back to the text

  4. Prompt — The input text sent to a model. For generative models it describes the desired result and strongly shapes the output.

    Back to the text

Frequently asked questions

What should you check first in this article?
Strata supports OpenAI Chat Completions and Anthropic Messages, plus the OpenAI Responses API starting in Strata 0.1.39. By default, the server accepts requests at `127.0.0.1:8080` and returns Qwen3.8-Flash-Next answers. A coding agent on the same PC uses the matching base URL; Codex CLI can connect through `/v1/responses`. Check `/health` and `/v1/models`, then send a short real POST request. Keep testing on loopback; do not change the host to `0.0.0.0` to expose it to a network.