Ask Options

View Source

Advanced options for fnord ask.

Basic usage

fnord ask -q "how does the auth middleware work?"
fnord ask -q "refactor the config module" --edit

Model tuning

--smart / -s

Use a more capable (and more expensive) model for the response. Trades speed and cost for improved accuracy on complex questions.

--reasoning / -R LEVEL

Set the AI's reasoning depth. Levels: minimal, low, medium, high.

Higher reasoning produces more thorough analysis but uses more tokens and takes longer.

--frippery / -V LEVEL

Set model verbosity. Levels: low, medium, high.

Controls how verbose or terse the AI's response style is.

Edit mode

--edit / -e

Permit the AI to create, modify, and delete files in the project. Without this flag, the AI operates in read-only mode.

--yes / -y

Auto-approve file edit prompts. Requires --edit. Can be repeated (-yy) to also auto-approve potentially dangerous operations.

In a git repository, --yes also skips the interactive post-session review for fnord-managed worktrees and attempts the usual merge-and-cleanup flow automatically. This does not apply to user-supplied worktrees via --worktree.

--auto-approve-after / -A SECONDS

Automatically approve pending edit prompts after SECONDS of no user input. Mutually exclusive with --auto-deny-after.

--auto-deny-after / -D SECONDS

Automatically deny pending edit prompts after SECONDS of no user input. Default: 180 seconds (3 minutes). Mutually exclusive with --auto-approve-after.

--worktree / -W PATH

Use PATH as the conversation's working directory for this run. When you resume the conversation later, fnord reuses that recorded worktree path instead of the current shell directory. PATH must be an existing directory. See Worktrees for the full worktree lifecycle.

Conversation management

--follow / -f UUID

Continue an existing conversation. The AI retains context from prior messages.

--fork / -F UUID

Create a new conversation branched from an existing one. The new conversation starts with a copy of the original's messages but diverges from the fork point.

--replay / -r

Replay a conversation to stdout (use with --follow).

Output

--save / -S

Save the AI's response to ~/fnord/outputs/<project>/<slug>.md.

--tee / -t FILE

Write a clean transcript of the conversation to FILE. Prompts before overwriting an existing file in interactive use; non-interactive runs fail and require --TEE.

--TEE / -T FILE

Like --tee, but truncates an existing file without prompting.

--quiet / -Q

Suppress normal UI output.

Environment variables

FNORD_SMART_MODEL

When set to true or 1, fnord uses the smart model on every invocation - equivalent to passing --smart/-s to every ask. Handy when you always want the heavier model and do not want to remember the flag. CLI --smart still wins when both are in play.

FNORD_TOOL_ROUND_CAP

Caps the number of tool-call rounds per coordinator completion. Defaults to 75. When the cap is hit, fnord withdraws the tool surface (no further tool calls are executed), injects a nudge telling the model to produce its final response, and lets the conversation finish. Lower the cap to force earlier wrap-up; raise it when a repo legitimately needs more exploration. Accepts any positive integer; malformed values fall back to the default.

FNORD_DEBUG_TOOL_CALLS

See Debugging Tool Call Errors.

FNORD_DEBUG_VALIDATION

Traces automatic validation decisions. See the same section in the asking-questions guide.

After each response, fnord prints:

  • Duration
  • Token usage and context window percentage
  • Conversation ID (with clipboard status when copying succeeds)
  • Active worktree path (if any)
  • Index staleness (new, changed, deleted files since last index)
  • Memory counts (session, project, global)