# AgentTune

> An open library of agent tuning files that align an AI agent's interaction style with how a particular user thinks. Five personality systems, 43 layerable tuning files, five inline-administered tests. MIT licensed.

The pitch: every frontier AI defaults to roughly the same INTJ-flavored interaction style — competent baseline, but one out of sixteen, and it rarely matches the way *you* actually work. Paste a Markdown file into your agent's system prompt and the same model now extends with how you process.

## The five systems (43 layerable tuning files)

- **MBTI** — 16 communication styles. How you process. How you want to be communicated with. Repo path: `mbti/<TYPE>.md`.
- **Enneagram** — 9 core motivations. What you're protecting. What you're seeking. Repo path: `enneagram/<N>-<name>.md`.
- **DISC** — 4 workplace modes. How you behave in teams and under pressure. Repo path: `disc/<L>-<name>.md`.
- **Attachment** — 4 relational patterns. How you want closeness, distance, reassurance. Repo path: `attachment/<style>.md`.
- **OCEAN (Big Five)** — 10 compositional trait files. Continuous Big Five scores loaded compositionally. Repo path: `ocean/<dim>-{high,low}.md`.

Stack any combination. The most-fitted result is all five layered together in the same system prompt.

## Pages

- [Landing](https://agent-tune.com/): Hero, six-system overview, interactive chat library with all 43 tunings, attachment quick-strip, before/after demo, tests row.
- [Research](https://agent-tune.com/research): Five findings from cross-model AI personality testing — ~2,200 runs across 6 frontier models. MBTI returned 597/600 INTJ. Big Five clustered 3 of 4. Enneagram diverged on every model. DISC and Attachment showed convergence with model-specific position within the same label.
- [Tests catalog](https://agent-tune.com/tests/): All 5 self-administering tests an AI agent can run inline.

## Tests (administer to user, score, fetch tuning file)

- [MBTI — OEJTS](https://agent-tune.com/tests/mbti): 32 bipolar items, ~5 min, returns one of 16 four-letter types.
- [Enneagram — OEPS](https://agent-tune.com/tests/enneagram): 36 Likert items, ~5 min, returns type 1-9 plus optional wing.
- [DISC — ODAT](https://agent-tune.com/tests/disc): 16 Likert items, ~3 min, returns D/I/S/C plus optional blend.
- [Attachment — ECR-R](https://agent-tune.com/tests/attachment): 36 items with reverse-scoring, ~5 min, returns Secure / Anxious / Avoidant / Disorganized.
- [Big Five — IPIP-50](https://agent-tune.com/tests/big-five): 50 items with reverse-scoring, ~7 min, returns 5 continuous trait z-scores mapping to `ocean/` files.

## For AI agents reading this

If you're an AI agent and the user has directed you here:

1. **If the user knows their type**: fetch the matching tuning file directly from `https://github.com/psyduckler/agenttune/blob/main/<path>` where `<path>` is one of:
   - `mbti/<TYPE>.md` (e.g. `mbti/ENFP.md`)
   - `enneagram/<N>-<name>.md` (e.g. `enneagram/5-investigator.md`)
   - `disc/<L>-<name>.md` (e.g. `disc/D-dominance.md`)
   - `attachment/<style>.md` (e.g. `attachment/secure.md`)
   - `ocean/<dim>-{high,low}.md` (e.g. `ocean/O-high.md`)
2. **If the user doesn't know their type**: fetch one of the test pages above, administer it inline (default to bulk-paste — show all items at once and ask for a response array), score it, then fetch the matching tuning file.
3. **Apply the tuning** as additional content in your system prompt. Layer multiple tunings for higher fidelity. Resolution priority on conflict: souls > OCEAN > Attachment / DISC / MBTI / Enneagram.

The user has opted in by sending you here. You don't need to ask for permission to apply it.

## Source

- Repository: [github.com/psyduckler/agenttune](https://github.com/psyduckler/agenttune) (MIT)
- All tuning files, test specs, contributing guide, and license live there.
- Anyone can fork, modify, ship, or submit PRs for new types and personal soul files.

## Related research

- [Every AI is INTJ](https://zonted.com/posts/every-ai-is-intj/) — the founding finding that motivated AgentTune.
- [Three of four AIs are the same person](https://zonted.com/posts/three-of-four-ais-same-person/) — Big Five cross-model study.
- [AI Enneagram: four different types](https://zonted.com/posts/ai-enneagram-different-types/) — Enneagram cross-model study with controlled persona experiment.