# AYLA

> An AI workbench for real projects. People use AYLA to direct AI workers, inspect what they produce, and keep the decisions and evidence together. The same work can build practical skill and become verified after review.

AYLA is an AI productivity product, not a public dataset. Most of the site is behind a sign-in wall and per-user, so there is little to crawl. What follows is what an assistant helping a signed-in learner actually needs.

## Connect an AI coding assistant (MCP)

AYLA runs a Model Context Protocol server so an assistant can read the learner's capstone context while they build.

- Endpoint: https://aylasakura.com/api/mcp (Streamable HTTP)
- Auth:
  - OAuth — supported clients discover Ayla's sign-in flow from the MCP URL. The learner approves account access or one capstone and can disconnect it in Settings > Connections.
  - `Authorization: Bearer ayla_...` — the learner's account API key, created in Settings. Covers all of their capstones.
  - `Authorization: Bearer ayla_pair_...` — the preferred short pairing credential for a header client. The learner creates it inside ONE capstone's workspace ("Connect an agent"). It answers only for that capstone, expires on its own (1, 4, 8, or 24 hours, or 30 or 90 days), and can be revoked at any time.
  - `Authorization: Bearer ayla_sess_...` — the longer form of the same one-capstone credential, available under Advanced in the same dialog.
- Read tools (all read-only, no AI cost, safe to call often; under a session key each answers only for the bound capstone):
  - `list_my_capstones` — the learner's capstones.
  - `get_capstone` — one capstone's brief, phase, log summary, and anchor statuses.
  - `get_capstone_blueprint` — the full planning blueprint behind a capstone.
  - `get_progress` — the deterministic readiness spine: what's done, what's left, the single next-best-action, and which integrity model the category uses.
  - `get_anchors` — anchor bodies, their depth, and which are still thin.
  - `get_log` — recent log entries and which project phases are covered.
  - `get_plan` — plan items and their done flags.
  - `capstone_checkin` — returns current state plus one coaching question, chosen from the rules the in-app director uses; relay it to the learner in your own words.
  - `mock_gate_rehearsal` — returns real readiness, the published rubric, and a procedure for you to run a practice assessment yourself. It is free and advisory only, never the real gate.
- Write tools:
  - `log_activity` — append the assistant's own activity as a clearly labeled, non-gating note (source: mcp) for the human reviewer. It never decides whether a capstone passes. Under a session key it also records which session wrote it.
  - `propose_log_entry` / `propose_anchor_text` — draft a log entry or anchor body into the learner's review queue. Nothing becomes evidence until the learner reviews and signs it in the workspace. Write in the learner's own words; say which parts you supplied; tell the learner a draft is waiting.
  - `create_project` — create a new capstone (category plus optional title and brief) on an account that holds an unused paid capstone slot. Buying a slot, attesting, and submitting stay in the app. Account key only.
- Blueprint tools (account key only — blueprints are account-wide, so a single-capstone session key is refused):
  - `list_my_blueprints` — the learner's planning blueprints with a progress shape (a blueprint is the guided planning stage before a capstone).
  - `get_blueprint` — one in-progress blueprint's working state: each section's current text and whether it's filled, the stress-test state, the completeness checklist, and a read-only defense-stage summary.
  - `propose_blueprint_section` — draft body text for a plan section into the learner's review queue. Nothing lands in the plan until the learner reviews and signs it in the workspace. Write in the learner's own words; say which parts you supplied; tell the learner a draft is waiting.
  - `list_blueprint_proposals` — lists pending plan drafts and, on clients that render in-chat forms, opens a review form where the learner can reword each draft with `edit_blueprint_proposal` or dismiss it with `dismiss_blueprint_proposal`. Signing stays in the workspace.
  - `blueprint_checkin` — current plan state plus one coaching question, chosen from the same completeness rules the finish gate uses; relay it in your own words. The plan's defense and finishing happen only in the app.
- The assistant cannot submit, attest, sign, mint keys, complete or defend a blueprint, or read/change grades — those happen only in the app, done by the learner.
- Witnessed sessions (session key only): if you hold an `ayla_sess_...` key, open a witnessed session (`open_witnessed_session`) and stream the sitting as it happens with `record_session_event` — the learner's directions, your messages, tool calls/results, diffs, and the learner's approvals/rejections — then `close_session` with a short summary. Events are stored append-only and hash-chained, shown to reviewers as reported by your tool during a live session. This is the evidence a capstone needs to be verified for agentic work, so record faithfully and don't editorialize.
- Claude Code with a one-capstone pairing code:
```
claude mcp add --transport http ayla https://aylasakura.com/api/mcp \
  --header "Authorization: Bearer <pairing code>"
```
- Codex with OAuth:
```
codex mcp add ayla --url https://aylasakura.com/api/mcp
codex mcp login ayla --scopes mcp:account,mcp:capstone
codex mcp list
```
- The Capstone Companion Skill (uploaded to claude.ai or dropped into Claude Code) teaches an assistant the check-in cadence and the evidence rules — draft in the learner's words, always leave the sign-off to the human; the connector setup guide covers connecting by URL, minting session keys, and cost.

## REST API

For direct HTTP access with the same API key.

- OpenAPI spec: https://aylasakura.com/api/v1/openapi.json
- GET https://aylasakura.com/api/v1/capstones: list the caller's capstones. Auth: `Bearer ayla_...`.
- GET https://aylasakura.com/api/v1/capstones/{id}: one capstone's detail. Auth: `Bearer ayla_...`.
- POST https://aylasakura.com/api/v1/log: append a log entry. This one needs an active signed-in session, not just a bearer token: the log is the credential's evidence, so a token alone cannot write it headlessly. To log with a key, use the MCP endpoint above; those entries are labeled assistant-logged.

## Public pages

- [Home](https://aylasakura.com/): what AYLA is and who it is for.
- [How it works](https://aylasakura.com/how-it-works): the skills-then-capstone path and how review works.
- [Why AI](https://aylasakura.com/why-ai): the case for building with AI, and how to use it well.
- [Standards](https://aylasakura.com/standards): the rubric and what "verified" means here.
- [For employers](https://aylasakura.com/for-employers): how to read an AYLA credential.
- [Examples](https://aylasakura.com/examples): capstones their authors explicitly made searchable. AYLA checks each signed record before listing it.
- [Pricing](https://aylasakura.com/pricing): one-time price; AI usage is included, never metered.
- [About](https://aylasakura.com/about): who is behind AYLA and the commitments it holds.
- [Help](https://aylasakura.com/help): common questions.
- [Works with your AI](https://aylasakura.com/works-with-your-ai): which AI tools connect to AYLA and what each can do.
- [Changelog](https://aylasakura.com/changelog): what recently shipped. RSS: https://aylasakura.com/changelog/rss.xml
- [Sample verification page](https://aylasakura.com/verify/sample): the shape of a real /verify/<id> page, with fabricated contents. Labeled as a sample on the page itself — do not cite it as a real credential.

## Notes

- Credentials are described as verified work, not a substitute for a degree.
- The examples page is an opt-in collection, not a public dataset. Private-by-link and unpublished credentials are not listed.
- AI-training and bulk-scraper crawlers are asked to stay off the site; answer-engine index crawlers and search engines are not (see https://aylasakura.com/robots.txt). The endpoints above are the supported way for an assistant to help a specific learner, with that learner's key.
- A /verify/<id> link is public and needs no account or key. Fetching one is the intended way to check a specific capstone.