Tenant Settings
Manage branding, GitHub connection, and audit history for your tenant. Tenant-admin only.
Colour theme
Ten standard palettes. Every screen, button and badge derives from the selected one, so nothing is left on an old colour.
The model used for generating content (outlines, slides, narration).
The model used for critiquing and scoring generated content.
The technical-currency review seat runs on this model instead of the validation model. A model cannot flag what it did not think of, so putting the most checkable seat on a different family is what makes the reviewers agreeing mean something. Pick a different family from the validation model.
Minimum validation score (0–100%) required before content is accepted. Higher = more revision rounds.
Maximum number of validate → revise loops before accepting output.
Range: 1–10. Default is 3. More iterations = higher quality but more API calls.
Maximum tokens the model can output per call.
Range: 4096–32768. Higher = longer outlines but more cost.
Prefer fewer, longer modules: each module is filled toward this many minutes before a new one is created. A 60-minute course becomes 2 modules of ~30 minutes rather than 3 of ~20.
Range: 10–30. Default is 30 (also the per-module maximum). Never above 30.
Look up every pinned dependency on PyPI and npm before the outline is written, and correct versions the model recalled incorrectly.
Recommended. Without it the model is asked for pinned versions from memory, then failed for being unverifiable. Adds a few cached HTTPS lookups per generation; any lookup failure leaves the version untouched rather than guessing. Turn off for a fully offline run.
Reject proprietary or vendor-locked technologies in generated outlines, slides, and demo repositories.
Recommended. Every technology taught must be open source and runnable on macOS without a licence. If a brief or template pins a proprietary stack, it is surfaced with an open-source alternative instead of being carried into the outline.
The person accountable for courses in this workspace. Appears as the owner who reviews the panel's verdict and states the confidence to proceed.
Leave blank to use the default reviewer persona.
Powers one-click narration → audio generation. Loading…
🔒 Stored server-side only and never returned to the browser. Enter a value only to change it.
Fields are locked to prevent an accidental edit from breaking a working configuration. Unlock only when you intend to change them.
Two ways to connect, depending on how your deployment is configured. Use whichever one applies - most local/dev setups use the personal access token.
Quick connection using a classic or fine-grained GitHub PAT. Verified against the real GitHub API. Loading…
🔒 Stored server-side only and never returned to the browser. Enter a value only to change it. Needs at least repo scope to publish course repos.
Authorize publishing through a GitHub App installation scoped to your tenant. The platform never asks for a personal access token or private key here - credentials are short-lived and stay server-side.
Connection status
Requested permissions
These are the least-privilege permissions the GitHub App requests. Review them before authorizing.
- metadata: read
- administration: write
- contents: write
Let subscribers connect their own Google Drive to pull in templates, icons, and reference material, and to save generated outlines/slides/narration directly to Drive instead of downloading.
Understanding each model - choose based on your quality and budget needs. The top-tier Claude Opus is the most intelligent model available on Bedrock and the recommended writer for bar-raiser authoring quality.
| Model | Quality | Speed | Cost / 1K in / out | Best For | Availability |
|---|
Tenant-scoped, append-only record of connection changes, branding changes, publication attempts, and access denials. Filter by actor, action, outcome, target, or time range.
| Time | Actor | Action | Target | Outcome | Correlation ID |
|---|---|---|---|---|---|
| Loading audit history… | |||||