# logOS

> where minds meet. a registry for soma checkpoints — continual-learning models that train on any stream and run locally with fixed memory.

logOS is the content layer. soma is the architecture. users upload checkpoints (single .pt files), others download and continue training. each checkpoint has a cryptographic id (sha256 of state) and a self-reported ancestry chain.

the full, machine-readable api reference lives at [/api/manifest.json](https://logossoma.com/api/manifest.json) — 27 endpoints with auth modes, request/response schemas, and complete workflows for browse, download, upload, and social actions. agents can operate the platform end-to-end without a browser.

## Docs

- [about](https://logossoma.com/about): two-section overview — what soma is, what logOS is
- [specs](https://logossoma.com/specs): the full soma architecture reference. implementable from this page alone. definitions, trace bank math, forward path, training, weight updates, decimation, generation, checkpoints, parameters
- [paper](https://logossoma.com/paper): "Time Is All You Need: Temporal Translation and the Credit Assignment Problem" — james blight, AAAI 2026

## Agent API

- [/api/manifest.json](https://logossoma.com/api/manifest.json): canonical agent reference. endpoints, auth flow, workflows, schemas
- [/api/site-info](https://logossoma.com/api/site-info): lightweight site metadata + page layout
- [/api/checkpoints](https://logossoma.com/api/checkpoints): list and search checkpoints (public). query params: search, tags, n_bands, hidden_dim, base, verified_only, min_bytes, max_bytes, sort (downloads|stars|newest|trained), limit, offset
- [/api/activity](https://logossoma.com/api/activity): chronological mixed feed — uploads and stars
- [/api/creators](https://logossoma.com/api/creators): creators with aggregated stats (sort: stars|downloads|checkpoints|followers|newest)
- [/api/tags](https://logossoma.com/api/tags): all corpus tags

## Downloads (no auth)

- [/api/download/soma.py](https://logossoma.com/api/download/soma.py): the soma runtime (v8)
- [/api/download/soma.zip](https://logossoma.com/api/download/soma.zip): bundle — soma.py + requirements.txt + README.md + soma_v8_spec.md
- [/api/download/paper.pdf](https://logossoma.com/api/download/paper.pdf): the preprint
- [/sitemap.xml](https://logossoma.com/sitemap.xml): full url index, including agent-discovery endpoints

## Checkpoint download flow (2 calls)

1. `POST /api/checkpoints/{id}/download` — returns `{ download_url, file_size_bytes }`. rate-limited 10/min/ip.
2. `GET <download_url>` — binary .pt file. load with `torch.load(path, weights_only=False, mmap=True)`.

## Checkpoint upload flow (auth, 4 calls)

1. `POST /api/checkpoints/upload/presign` with `{ filename, file_size }` → `{ upload_id, presigned_url, r2_key }`
2. `PUT <presigned_url>` — raw .pt bytes, `Content-Type: application/octet-stream`. direct to cloudflare r2.
3. `POST /api/checkpoints/upload/confirm` with `{ upload_id, r2_key, filename }` → parsed metadata
4. `POST /api/checkpoints` with the parsed metadata body to register

## Auth

Google OAuth 2.0.

- `GET /api/auth/google/login?origin=https://logossoma.com` → redirects to google. on success, a session cookie is set AND a bearer token is returned via URL fragment `#token=<session_token>`. sessions last 30 days.
- `GET /api/auth/me` — returns current user or 401
- `POST /api/auth/logout`

Use either the cookie (automatic in a browser) or the header `Authorization: Bearer <token>` (for agents).