Run ConfPlanner against a real model¶
You want the agent to call a live LLM and build a real schedule. The build and tests are green without any key (they mock the LLM); a key is needed only to run against a real model.
1. Set a provider key¶
Edit .env and set one key. .env is git-ignored — never commit a real key.
OPENAI_API_KEY=sk-... # default; application.yml routes to OpenAI models
# or:
ANTHROPIC_API_KEY=sk-ant-...
Both provider starters are always on the classpath, and .env is loaded into the app at startup
(via spring-dotenv), so the key you set here is picked up automatically — there is nothing else to
configure. If you set the Anthropic key, also point the model roles at Anthropic models in
application.yml — see Route models by role.
2. Start the shell¶
3. Invoke a goal¶
In the shell, type a free-text request after x (alias for execute); the platform matches it to a
goal:
To read the plan rather than just the answer, add prompt (-p) and result (-r) verbosity:
This is the same world-state a trace shows — see Enable Zipkin tracing for the visual version.
4. (Optional) Invoke a specific goal directly¶
To call one goal by result type instead of letting the platform match, use the bundled plan
command, which invokes PersonalSchedule via AgentInvocation:
If you have no key¶
Run the deterministic mock instead — see Run a deterministic demo with mock mode.
For the full list of shell commands and flags, see the CLI reference. For role names and their default models, see the model-routing reference.