Example threads. These questions and answers were written by the SDODS team to document real problems and how to fix them. The people named are illustrative, not real users. Questions asked on this site are marked new.
New to this. sdods doctor gives me a tidy screen and then a token section that lists what looks like six different ways to pay for the same thing.
ANTHROPIC_API_KEY, a claude CLI login, a codex CLI login, OPENAI_API_KEY with a base URL, local models through ollama, and SDODS_TOKEN. Do I need all of them? Any of them?
Accepted answer
You need zero of them to use SDODS, and exactly one of the first five to use the agent roles. The docs call that group "one-of" and the doctor prints them together for that reason.
ANTHROPIC_API_KEY — the claude adapter. Billed per token by Anthropic. The row's own hintsays it: or log in to Claude Code (claude login) — no key needed.
npm i -g @anthropic-ai/claude-code && claude login. No key; yoursubscription pays.
OPENAI_API_KEY (+ OPENAI_BASE_URL) — any OpenAI-compatible endpoint, including your own.ollama serve && ollama pull qwen2.5-coder:7b. No key, no bill, nothingleaves the machine.
SDODS_TOKEN (+ SDODS_SERVER_URL) is not in that group at all. It is SDODS's own token, it is free and self-issued, and it is only needed to talk to a hosted SDODS server — MCP over HTTP, or uploading run results. It never buys you a model.
Running suites, screenshots, lint, analyze, the dashboard and the stdio MCP server need no credential whatsoever.
And if you set more than one, the order that decides is fixed: ANTHROPIC_API_KEY → claude CLI → codex CLI → OPENAI_API_KEY → a local Ollama → fake. SDODS prints which one it picked at the start of a job.
Which trips people up when they add a key for one experiment and then wonder why their local model stopped being used. --adapter on the command, or agents.provider in the project yaml, beats the detection.
Wrap commands and output in triple backticks and they will render as a code block.
Checking whether you’re signed in…
Have a different question? Ask it.