Skip to content
Axiom Wiki

query

Terminal window
axiom-wiki query

Opens an interactive chat where you can ask questions about your wiki content. The agent searches relevant pages, synthesizes an answer, and cites sources.

You can also type questions directly in the interactive shell without the /query prefix:

> What did Alan Turing say about intelligence?

After answering, the agent offers to file the answer as an analysis page in wiki/pages/analyses/.

Section titled “Analyses-first search”

The agent checks wiki/pages/analyses/ first to see if the question has already been answered. Previously filed analysis pages receive a 1.5x boost in search ranking, so they surface ahead of raw entity or concept pages. This means repeat questions are answered faster and more consistently.

Section titled “Structural tree search”

For large wikis with thousands of pages, the query agent avoids dumping the entire catalog into context. Instead, it navigates structurally:

  • Inspects high-level category and tag trees using list_pages({ mode: "tree" }).
  • Runs targeted hybrid searches with section-level anchor resolution.
  • Explores graph connections and backlinks (get_backlinks) to trace relationships between concepts and entities.

Strict Forensic Grounded Mode (Zero Trained Knowledge)

Section titled “Strict Forensic Grounded Mode (Zero Trained Knowledge)”

To enforce strict, zero-hallucination answers where the AI is strictly barred from using pre-trained parametric knowledge:

Terminal window
# Launch interactive chat in forensic grounded mode:
axiom-wiki query --forensic
# Or using the -g alias:
axiom-wiki query -g
# Or pass a question directly:
axiom-wiki query --forensic "What was the initial revenue in paper 1?"

Inside the interactive shell or Query screen:

  • Type /forensic [question] or press Ctrl+G to toggle Strict Grounded Mode on and off.
  • The UI displays a 🛡️ STRICT GROUNDED (Zero trained knowledge) status banner.

Grounding Rules Enforced:

Section titled “Grounding Rules Enforced:”
  1. Absolute Pre-Training Ban: The model is forbidden from using outside world knowledge or guessing facts not explicitly in the wiki.
  2. Mandatory Inline Proof: Every claim must directly cite a retrieved wiki page or source locator: (→ [[page]], source: doc#locator).
  3. Refusal on Insufficient Evidence: If the retrieved documents do not contain the answer, the model explicitly states: “The wiki does not contain sufficient information to answer this question.” and states what sources would be needed. Never guesses from memory.