What you’ll learn
- When to use direct action runtime calls instead of long-lived agent loops
- How the core action families map to practical workflows
- How to keep a runtime demo deterministic without external API keys or network access
-
How to swap fixture-backed families to
Jido.AI.Actions.*modules in your own app
Covered action families
-
LLM envelopes: fixture-backed
chat,complete, andgenerate_object -
Tool calling: shipped
list_tools/execute_toolplus one deterministiccall_with_toolscompanion -
Planning: deterministic
plan,decompose, andprioritize -
Reasoning: deterministic
analyze,infer,explain, andrun_strategy -
Retrieval: shipped
upsert_memory,recall_memory, andclear_memory -
Quota: shipped
get_statusandreset
How this demo stays deterministic
This page runs real Jido.Exec.run/3 calls on every button press.
-
Retrieval and quota use the shipped
Jido.AI.Actions.*modules directly. - Tool discovery and direct tool execution use the shipped tool-calling actions directly.
- The LLM-backed families use local fixture actions in this repo so the site demo never depends on provider credentials.
That keeps the public example truthful: the runtime surface is real, the code is local, and the outputs stay repeatable.
Pull the pattern into your own app
In your own project, keep the same Jido.Exec.run/3 shape and swap the fixture-backed families to the production modules from Jido.AI.Actions.*.
- Keep retrieval and quota as-is if the in-process stores fit your needs.
- Replace the fixture tool with your own tool Action modules.
-
Replace the fixture LLM/planning/reasoning actions with the shipped
Jido.AI.Actions.*modules once provider credentials are configured.