Getting Started¶
This guide walks a new user through Polaris end to end: from an empty brain to a published, cited knowledge base you can consult and keep fresh. If you are an operator standing up the stack, start with the Dev Loop instead — this page is about using the app.
What Polaris is¶
Polaris is a multi-tenant Brain Builder. Each tenant operates one Brain — a governed knowledge base that is grown, reviewed, and published through a strict pipeline:
flowchart LR
E[empty] --> B[building] --> P[published]Four AI agents do the work, but a human approves every change that reaches the published corpus. Two agents propose work, one curates and publishes it, and one serves answers from what has been published.
-
Scout — discovers new sources on the web.
-
Librarian — the single write gate: review → certify → publish.
-
Advisor — answers questions from the published corpus, always cited.
-
Sentinel — watches known sources for drift and staleness.
The end-to-end lifecycle¶
Follow this loop the first time you build a brain. Each step links to its full guide.
1. Discover — find sources with Scout¶
Open Scout ( in the left rail). Set a discovery policy (your goals, jurisdictions, topics, and which domains are allowed), then click Run Scout. Scout searches the web for net-new sources and returns a ranked candidate queue.
Review the candidates and Target the good ones. Targeting hands each document to the Librarian — Scout never writes to the corpus itself. → Scout guide
No web search configured?
You can also add documents manually from the Workspace or the Librarian's Add document button — the same review pipeline applies.
2. Curate — review and approve in the Librarian¶
Open Librarian (). Each targeted or uploaded document is extracted in the background into a change set — a bundle of proposed facts, rubrics, and graph entities. Change sets land in the Awaiting review queue.
Open a change set to review it: include or exclude individual knowledge units, resolve any escalations, then approve (or reject). This human-in-the-loop (HITL) gate is the heart of Polaris. → Librarian guide
3. Certify & publish¶
When the draft looks right, click Certify & publish. Certification runs a safety hard-gate; if it passes, a new immutable version is published. Publishing also generates the human-readable Wiki. → Librarian guide
4. Consult — ask the Advisor¶
Open Advisor (). Run a verb — answer, assess, redline, or retrieve — against the published corpus. Every result is grounded in source passages (cite-then-verify) and shows its citations and retrieval trace. → Advisor guide
5. Steward — refine the brain from the wiki¶
Reading an article and spot something to fix? From the Wiki a Curator can act on it in place: Edit a description, Flag for removal to prune a concept, or Propose as enrichment to promote a grounded Deepen synopsis into the article. Each creates a proposal that lands in the Librarian's Wiki proposals queue — confirm it there, then certify & publish to make it live. It's the same governed loop as ingestion, just started from the finished artifact instead of a document. → Wiki stewardship
6. Monitor — keep it fresh with Sentinel¶
Open Sentinel (). Run a sweep to re-check already-ingested sources for freshness, date transitions, broken grounding, and conflicts. Sentinel raises flags — confirm one and it routes the affected source back into the Librarian loop for re-curation. Sentinel is the system-initiated twin of wiki stewardship (step 5) — both propose into the same governed loop. → Sentinel guide
What you can do depends on your role¶
Polaris is role-gated. In increasing capability:
| Role | Can typically… |
|---|---|
| Reader | View dashboards, the Wiki, and read-only panels; consult the Advisor. |
| Curator | All of the above, plus run Scout/Sentinel, ingest, review change sets, steward the wiki, certify. |
| Admin | All of the above, plus manage settings, API keys, team, and cost. |
Actions you cannot perform appear locked (dimmed with a lock icon) or read-only. See Concepts & Roles for detail.
Where to next¶
-
The vocabulary — brains, change sets, candidates, flags, versions — and who can do what.
-
The left rail, agent dashboards vs. chat, and how to move around.