First Run (Minimal + LLM)¶
Goal¶
Get two runs working end-to-end:
- a minimal run (no model call) to validate the kernel wiring
- an LLM-backed ReAct run to validate model + parser + tools + trace
0) (One-time) configure model for LLM examples¶
QitOS examples read:
OPENAI_BASE_URL(provider endpoint)OPENAI_API_KEY(orQITOS_API_KEY)
Fastest setup:
If you prefer CLI-only config, see: Configuration & API Keys.
1) Run the minimal agent (no LLM)¶
What to check:
- The program completes with a final result.
- The run has a clear stop reason.
- If traces are enabled, you can find a run folder under
runs/.
2) Run an LLM-backed agent (ReAct)¶
This run exercises the default model path:
AgentModule.decide(...) -> None- Engine builds messages (system + memory + prepared user text)
- Engine calls
llm(messages) - Parser turns text into
Decision(Action(...)) - Engine executes the tool call and reduces into state
What to check:
- You can see model output in the terminal render (unless
--disable-render). - Tools are actually called (you should see action results in scratchpad/trace).
- A run folder exists under
runs/and containsmanifest.json,events.jsonl,steps.jsonl.
Next¶
- If model calls fail: Configuration & API Keys
- Inspect your run with qita: qita Guide