Slancha decision setup — decision-public-setup-v1

Help me use the currently available Slancha decision gateway at
https://decide.slancha.ai. Explain the steps before acting.

1. Read public GET /api/pipelines and report actual service availability.
   Open https://decide.slancha.ai/static/workspace.html. I must connect with
   my existing operator-approved service key myself. Never ask for a key in
   chat, read browser auth/key storage, or put credentials in prompts or URLs.
   If I lack access, help me request access/costs at https://slancha.ai/contact.
   Do not create keys, grant permissions or change credentials/authentication.
2. Agree a supported decision type, scrubbed labeled data and cost scope with
   me before uploading or training. Image/audio projects fit classifier heads
   on frozen features; trace models use text/JSON features; tables use numeric
   context. Text creation is operator-assisted outside this workspace.
3. Use the existing Build/Projects/Try/API tabs. Data needs 2–64 labels and at
   least 3 examples per label; at least 4 per label for a separate result check.
   The gateway accepts at most 1,000 examples per upload and 64 MB per request.
   Media integrations accept inline base64/data URLs only, never local paths
   or arbitrary media fetches. Get my explicit authorization for uploads,
   training or deployment; do not run paid workloads just to check setup.
4. The first ready version deploys automatically. Observe actual queued,
   training, failed or ready status. A failed run is not a successful model.
   Check held-out results without claiming real-world accuracy/calibration.
5. Test an existing deployed version with a safe synthetic example and saving
   corrections off. Confirm an actual response before claiming setup worked.
   An uncertain upload/train/deploy response requires reading current model
   state before any manually approved retry. Never automatically replay writes.
6. Use the existing API tab for the owned model identifier and request shape.
   Authentication is an existing service key in the Authorization header, read
   only from an approved secret mechanism in server software. Never embed it
   in frontend code, logs, analytics, a repository, chat or a URL. Project
   inference stores input by default unless the integration sends store:false.

Current gateway routes (not the upcoming account adapter):
GET /api/projects; POST /api/projects with name and supported modality.
GET /api/projects/{id}; POST /api/projects/{id}/examples;
POST /api/projects/{id}/train; POST /api/projects/{id}/deploy with version.
POST /v1/systemone with the owned proj-{id} model and trained-label questions;
include store:false to avoid retaining project inference inputs.
GET /api/projects/{id}/export downloads dataset JSONL, not model weights.

This service is hosted on Slancha-operated servers.
This list covers decide.slancha.ai only.
Subprocessors for this service:
- AWS CloudFront (content delivery and proxy, global edge)
- Cloudflare (secure tunnel and network transport)
- Cloudflare R2 (file storage)
This list may change; we notify customers by a dated update to this page.
Last updated: October 14, 2026
No tracing is enabled by setup. Account-based model access, self-serve
signup/billing and model artifact downloads are not available in this release.
Spoken-conversation availability comes from service state; do not assume it is available.

Clipboard/load failure: reload or select/copy the recipe manually. Revoked or
missing access: stop and ask Slancha; do not fall back to a shared admin key.
Uncertain action: read state first. No successful request means setup unfinished.