Give every AI agent your defaults
Datailor turns your recurring corrections into local, editable preferences that every agent can use.
pipx install git+https://github.com/fyaic/bondie-preference-MCP.git
Our Stance
Low-Effort Maintenance
No rule gardening. No config chores. Datailor turns repeated corrections into editable preferences as you work.
Low-Interruption
Your defaults should run quietly in the background. Datailor reduces repeated prompts without hiding control.
Anti-Lock-in
Your preferences live in a Markdown file on your machine. Take them anywhere. Leave anytime.
Anti-Black-Box
Every preference is visible, editable, and traceable. Inspect what was learned, why it applied, and what changed.
Self-Growing
Every correction becomes signal. Your preference layer compounds across agents instead of resetting in every chat.
Local-First
No cloud. No API calls. No data leaves your device unless you choose.
Agent-Neutral
Codex, Claude, Kimi, Cursor, Trae: the preference layer belongs to you, not to one assistant.
Human-Editable
Everything is plain text. Review, rewrite, delete, or promote preferences without learning a new system.
Context-Aware
Preferences are matched to the current task. Stale, conflicting, or irrelevant rules are not blindly injected.
Conflict-Safe
When preferences disagree, Datailor pauses automation and surfaces the conflict instead of guessing silently.
The agent is not lacking intelligence. It is missing your defaults: when to test, how to write, what to avoid, when to ask, and when to proceed quietly.
Datailor turns those repeated corrections into a local preference layer, so your standards travel with every agent instead of being re-explained in every chat.
DATA, DETAIL, AI, TAILOR
Auto-Learning, Zero Maintenance
Scans local AI chat history (Kimi, Codex, Claude, Cursor, Trae). Extracts recurring preferences. Stores as human-readable Markdown. You correct once, it remembers forever.
Decision-Making at the Right Moment
Five lifecycle hooks: session start, message arrival, AI reply, action execution, session end. Dynamically calculates preference validity. Expired or conflicting preferences are never blindly injected.
Full Transparency
Local UI panel: Executive Summary, Preference List, Conflict View, Injection Log, Evolution Board. Everything traceable. Nothing hidden.
Evolves from Feedback
Confirm → "Keep this." Reject → "Remove this." Correct → "Rephrase this." Auto-snapshot before every change. Rollback anytime.
Three Steps
One-command scan. Auto-extracts preferences from your history.
Open the local panel. See what it learned.
Next time AI asks, Datailor answers for you.
Privacy & Data
- Storage Human-readable Markdown. Editable. Portable.
- Offline Works without network.
- Privacy Zero cloud APIs unless you configure a model backend.
- Snapshots Auto-backup before every change.
Why Not...?
| Approach | The Problem | Datailor's Way |
|---|---|---|
| Skill | "Enable this Skill?" — confirmation loops | Pure MCP tool. Silent invocation. Zero friction. |
| Cloud Service | Data leaves your device. Platform lock-in. | Everything local. Pure Markdown. Export anytime. |
| Manual Config | Writing rules is anti-human. Abandoned in weeks. | Fully automatic. Zero maintenance cost. |
| Keyword Matching | Can't understand semantics. High false positives. | Semantic recall + behavioral patterns + dialogue structure. Multi-route fusion. |
What's Next
- PyPI / pipx distribution
- Codex / Kimi / Claude / OpenClaw config generators
- Compliance checking on AI output
- Background file watching
- Vector indexing
- Obsidian plugin