Diagnostics

Kallamo can show, under each reply, everything that went into it. Turn the parts you want on in Settings › Diagnostics. Nothing here changes the reply; it only shows how it was made.

Option	What it shows
Retrieval and context	What was found, what made it in, perspective scopes, and the exact context sent.
Agentic retrieval trace	For profiles with smart search: the searches, the results, why it stopped and what it cost. Never the model’s private reasoning.
Narrative State	For roleplay turns: the triage, the deep analysis, proposals, retries and timing, and the exact narrative context the narrator received.
Token usage	Estimated tokens for the reply, retrieval and Narrative State, each counted once.

Reading them

• Answer missing a fact? Open Retrieval and context. If the fact was not found, see the tuning tips in How Kallamo finds context. If it was found but left out, the budget was full: see Context budget.
• Smart search stopped early? The trace says why: the turn limit was reached, the budget was reached, or the planner had enough.
• A roleplay record looks wrong? The Narrative State panel shows what the triage decided the turn touched, what the analysis recorded, and the raw record. “Nothing lasting happened” is a valid outcome for a turn of small talk.
• Costs higher than expected? Token usage splits the reply, retrieval and Narrative State, and shows what was read from the provider’s cache where it reports it.

The narrator’s plan

In roleplay, the narrator writes a private scene plan before each reply: who is in the scene, what each one can say, and what must stay unsaid. It never reaches the story. You can read it in the message’s diagnostics.