The AI That Trains Your Judgment
Every day you make dozens of decisions.
Some are trivial.
Some are emotional.
A few quietly shape your future.
We rarely track them.
We rarely measure them.
We almost never learn from them.
Decidr changes that.
Decidr is a local-first AI decision intelligence system.
It turns daily decisions into structured, reviewable experiments.
You:
- Log a decision (1–2 lines)
- Record your confidence level
- Let AI forecast possible outcomes
- Review results later
- Analyze your prediction accuracy and bias patterns
Over time, Decidr helps you improve judgment through feedback loops.
Most productivity tools track tasks.
Decidr tracks thinking quality.
It doesn't ask: "Did you finish it?"
It asks: "Was your reasoning accurate?"
For every logged decision, Decidr:
- Generates probability-weighted outcome predictions
- Surfaces key assumptions
- Suggests a review horizon
- Tracks user confidence
- Compares forecast vs actual outcome
- Detects long-term cognitive bias patterns
This creates a personal "decision accuracy dataset."
Over time, Decidr helps identify:
- Overconfidence patterns
- Risk avoidance tendencies
- Emotional decision clusters
- Timeline underestimation bias
- Category-based judgment variance (work, health, finance, etc.)
You build a personal cognitive fingerprint.
- Local-first storage — no data leaves your machine by default
- Designed for privacy-sensitive decision logging
- Structured SQLite storage
- AI layer can run via local models (default) or optional "bring your own key" for OpenAI-compatible APIs
Your decisions are your intellectual history.
Improving decision quality by even a small margin compounds:
- Better project execution
- Stronger investments
- Healthier communication
- Reduced reactive behavior
- More strategic long-term thinking
Decidr is not a task manager.
It's a judgment training system.
In a world obsessed with output optimization,
Decidr focuses on decision optimization.
And in the long run,
thinking quality wins.
pnpm install
pnpm tauri devBuild for production:
pnpm tauri buildIcons (required for build): place your app icon as app-icon.png in the project root, then run:
pnpm tauri icon app-icon.pngThis generates the icons in src-tauri/icons/.
- Decisions: Quick-add one line + optional context; category, confidence slider, horizon (days), tags. Save in < 2 seconds; Enter submits; ⌘K / Ctrl+K opens command palette.
- AI Prediction card: After saving, see 3–5 outcomes with probabilities (sum 100%), key assumptions, and a recommended next action. Stored for reproducibility.
- Review Due: List decisions whose review date is today or past and not yet resolved. Mark outcome (free text + good/neutral/bad + notes). App computes accuracy and calibration scores.
- Insights: Overconfidence (high confidence, low accuracy), category-level accuracy charts (local aggregations, no external analytics).
- Settings: Toggle predictor: Local vs OpenAI-compatible. For OpenAI: base URL, API key, model, temperature. API key is stored via the app store (app data directory).
- decisions: id, created_at, decision_text, context, category, confidence, horizon_days, review_date, tags (JSON), status (open | due | resolved).
- predictions: id, decision_id, created_at, model, prompt, raw_response, outcomes_json, assumptions_json, next_action.
- outcomes: id, decision_id, resolved_at, happened_outcome, outcome_rating (good | neutral | bad), notes, accuracy_score, calibration_score.
Scoring is deterministic and offline: accuracy from fuzzy match of happened vs predicted outcomes; calibration from Brier-style score on probability buckets.
- Rust:
cd src-tauri && cargo build/cargo test - Frontend:
pnpm dev(Vite on port 5173) - Full app:
pnpm tauri dev
- Run app:
pnpm install && pnpm tauri dev. Window opens with sidebar (Decisions, Review Due, Insights, Settings). - Decisions: Type a one-line decision, optional context, set category/confidence/horizon/tags, click "Save decision" or press Enter. Decision appears in list; select it and confirm the "AI Prediction" card appears with outcomes and next action.
- Command palette: Press ⌘K (Mac) or Ctrl+K (Windows/Linux). Navigate with arrows, Enter to switch page.
- Review Due: Create a decision with horizon 0 or 1 day so review date is today/tomorrow; go to "Review Due", select it, fill "What actually happened?", choose rating, click "Mark resolved". Check that accuracy/calibration appear.
- Insights: After resolving a few decisions (especially with high confidence), open Insights and confirm "Overconfidence" and "Category-level accuracy" sections appear when data exists.
- Settings: Switch to "OpenAI-compatible API", enter base URL and API key (or leave blank and save). Return to Decisions and add a decision; prediction should use the API if key is set.
- Rust tests:
cd src-tauri && cargo test— scoring (accuracy/calibration), probability normalization, and JSON validation should pass.