Skip to main content
FireConnect routes DeepSeek Harness (dsh) through Fireworks AI models. See the FireConnect overview for install and CLI basics.
Change models: fireconnect deepseek on --model <id>. See Models.

​
Prerequisites

​
Enable Fireworks routing

Or pass the key once:
Restart dsh after on or off if it is already running.

​
Default model

DeepSeek Harness routes a single default model. The default is kimi-fast-latest.

​
What gets written

FireConnect updates two files under $DSH_HOME (default ~/.dsh):
  • Adds a custom fireworks provider under llm-pi-ai.providers in settings.yaml, using the OpenAI-compatible endpoint at https://api.fireworks.ai/inference/v1
  • Sets agent-default-model to the selected Fireworks model
  • Stores the key as FIREWORKS_API_KEY in .credentials.yaml with file mode 0600
FireConnect snapshots both files under ~/.fireconnect/deepseek/ before the first change. Running fireconnect deepseek off restores them.

​
Browsing and picking models

Fire Pass keys only list Fire Pass routers. FireConnect rejects --model firerouter with Fire Pass (fpk_...) on every harness; use an fw_... key.

​
FireRouter

DeepSeek Harness cannot attach a local Anthropic key, so Anthropic pass-through requires workspace BYOK.

​
CLI reference

Run fireconnect deepseek help for all options.

​
Switch models

​
Turn off Fireworks routing

This restores the previous settings.yaml and .credentials.yaml snapshots.

​
Use a non-default config file

The credentials file remains beside the selected settings file.
Fireworks on Microsoft Foundry is not supported for DeepSeek Harness. fireconnect deepseek on uses the direct Fireworks gateway.

​
Source

FireConnect is open source: github.com/fw-ai/fireconnect