DeepSeek API¶
DeepSeek is a hosted AI provider, not a coding-agent CLI — there is nothing to launch and
no jsat connect deepseek/jsat deepseek command. It configures the AI backend JSAT's own
LLM-backed tools use (crack, prompt_optimize, security_review, and so on), the same way
anthropic/openai/gemini do.
DeepSeek's API is OpenAI-compatible, so JSAT reaches it through the same openai_compat
backend used for Gemini and LM Studio — jsat ai use deepseek just presets the base URL and
key environment variable for you.
Set API key¶
Get a key from platform.deepseek.com:
Activate¶
DeepSeek needs an explicit model — JSAT never guesses one:
jsat ai use deepseek --model deepseek-chat
# or the reasoning-focused model:
jsat ai use deepseek --model deepseek-reasoner
This sets the provider to openai_compat with base_url pointed at:
and api_key_env: DEEPSEEK_API_KEY, so DeepSeekProvider-style calls read the right key
automatically — no manual api_key_env edit needed in .jsat/config.yaml.
# Apply globally (all projects on this machine):
jsat ai use deepseek --model deepseek-chat --global
Verify¶
Inside the shell¶
Troubleshooting¶
- A
404/model-not-found style error fromjsat ai testmeans the model name is wrong — DeepSeek's own docs list the current model IDs;deepseek-chatanddeepseek-reasonerare the two general-purpose options at time of writing. - If
DEEPSEEK_API_KEYis unset,jsat ai use deepseekprints a warning at setup time; a missing/invalid key surfaces as an authentication error fromjsat ai test. - This is unrelated to "DeepSeek Harness" (
dsh), a separate open-source coding-agent CLI launchable viaollama launch dsh— JSAT does not currently wire MCP tools into that harness.