You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
See whether ClawDEA is saving you money
Estimated Savings / Cost in the Cost Control panel. A new section estimates the net dollar effect of ClawDEA's knowledge layer and IDE integration versus a standard Claude Code setup — the wiki librarian keeping exploration out of the main context, the IDE index replacing file reads — netted against what the layer costs to run (wiki/workspace upkeep and the per-turn primer). Shows this chat, this month, and all-time, each with a confidence range.
Honest by design. The estimate is a directional comparison against a baseline we don't actually run, and it can go negative; the section title flips to "ClawDEA Estimated Cost" when the all-time net is negative. Estimated lines carry a ± range; measured costs do not.
Refines on resume. Live sessions show a conservative floor; reopening a conversation reconstructs the full session from its transcript for a sharper estimate.
Spend tokens on completions only when you ask (#146)
Manual completion hotkey. A new Trigger Inline Completion action (default Alt+\, ⌥\ on macOS) requests a suggestion on demand — it always works, even with automatic completions on. Rebind it in Settings → Keymap → "Trigger Inline Completion".
Optional manual-only mode. Enable "Only request completions on hotkey" in Settings to suppress as-you-type completions entirely, so API tokens are spent only when you press the hotkey. Off by default — automatic completions are unchanged unless you opt in.
Fixes and polish
The activity hint stays put during long and parallel work. The pulsing "Thinking" dots no longer vanish mid-turn — they are re-asserted on each message and tool result and kept animating through long, quiet stretches, so a slow or parallel sub-agent run always shows that something is happening.
Chat view recovers after sleep or a display change (#36). When the machine wakes from sleep or you (un)plug a monitor or move the window between displays, the chat could freeze or render at half resolution and never recover. ClawDEA now rebuilds the view automatically (and /refresh-view forces it), replaying the conversation into a fresh, responsive surface.
Cost Control numbers reconcile and always show. The savings section now adds up — estimated per-lever savings, measured upkeep, and a single "Overall" bottom line — and the all-projects, all-time aggregates show immediately on a brand-new chat.
IDE index-tools savings are no longer stuck at $0.00. The first file read that the IDE index avoids is now priced at the full input rate, so the estimated saving reflects the real cost avoided.