Saberra Settings Guide

This guide is a tour of what you can configure in Saberra and, just as important, how fast each change takes effect. The headline: almost everything lives in the Control Center tab or in Hub Settings, and almost nothing requires a redeploy. (Members see an Account tab instead, which holds their own preferences only.)

Governing Purpose Statement (GPS)

The GPS is your organization's stated purpose. Sera uses it to score decisions for purpose alignment, so that governance-relevant decisions carry a note on how well they serve why your organization exists.

You edit the GPS directly in the Control Center. There is no redeploy: a change takes effect within about 30 minutes, after which new decisions are scored against the updated statement.

Hub Settings

Hub Settings are stored on a Notion page and override the environment defaults for your instance. Because they are read at the start of each poll cycle, a change takes effect within one 3-minute poll cycle, with no redeploy. Four settings live here.

Output Language: the language all extracted field values are written in. Set it to your working language and Sera will write summaries, task descriptions, decisions, and every other extracted field in that language.

Correction Mode: controls how the language-normalization scan handles records it finds written in the wrong language.

• A = Recommend Only — Sera flags but does not change anything.
• B = Propose an update in the Memory Review Queue — the default. Sera proposes the correction as a review item for a human to approve.
• C = Auto-Update Low-Risk — Sera applies low-risk corrections automatically and proposes the rest.
• D = Auto-Update All Allowed — Sera applies every allowed correction automatically.

Extraction Granularity: the default density of extraction.

• essential — only the most important records.
• standard — a balanced default.
• full — the richest extraction; full also enables the Interactions CRM auto-log, so every processed email and meeting is logged as a contact interaction.

Review dials: one dial per reviewable record type, set to auto, tiered, or always-human. This is how you decide what may promote itself after an objection window and what always waits for a person. Canon changes, policies, sensitive review, and the governance ledger are hard gates and are never affected by any dial value. The trust dial article covers how to choose.

Extraction Addendum

The Extraction Addendum is a short piece of org-specific context appended to Sera's extraction prompt. Use it to teach Sera what your organization is, so extractions come out sharper. For example: "This is a housing cooperative focused on land stewardship." It does not change what Sera can do; it changes how well she understands your world.

Email Notifications

Three things are configured here.

• The admin notification address: where Sera sends access requests and review alerts. This address must differ from the capture inbox (for example, capture@yourorg.com), so that notifications do not loop back into the ingestion pipeline.
• An outbound-email on/off switch: a master control for whether Sera sends any outbound email at all. Turn it off to run silently, for example during setup or testing.
• The Monthly Story stream: whether the monthly narrative is distributed by email, and to whom. It is publish-gated: nothing is sent until a person publishes the draft, so switching the stream on does not make Sera email anybody by herself.

System Configuration (read-only)

The System Configuration panel shows operational values you can see but not change from the dashboard: the AI model in use, the poll interval, and the tenant identifier. These are surfaced for transparency and troubleshooting. They are set at the infrastructure level, not in the dashboard, so treat this panel as a status readout rather than a control.

The takeaway

The settings that shape day-to-day behavior (purpose, language, correction mode, granularity, review dials, org context, and notifications) are all reachable without engineering help and take effect within minutes. The GPS refreshes within about 30 minutes; Hub Settings refresh within one 3-minute poll cycle. Only the read-only System Configuration values are fixed outside the dashboard. In practice this means you can tune Saberra to your organization confidently and quickly, and see the effect the same day.

Key points

Admin settings live in the Control Center tab, or in Hub Settings, and never require a redeploy. The Governing Purpose Statement drives decision purpose-alignment scoring and takes effect within about 30 minutes. Hub Settings are stored on a Notion page, override environment defaults, and take effect within one 3-minute poll cycle: Output Language (the language all extracted field values are written in), Correction Mode for language normalization (A = Recommend Only, B = Propose an update in the Memory Review Queue and the default, C = Auto-Update Low-Risk, D = Auto-Update All Allowed), Extraction Granularity (essential, standard, full, where full also enables the Interactions CRM auto-log), and the per-type review dials (auto, tiered, always-human) that decide what may promote itself. The Extraction Addendum appends org-specific context to Sera extraction prompt. Email Notifications set the admin notification address (must differ from the capture inbox), an outbound-email on/off switch, and the Monthly Story stream that distributes the story only after a human publishes it. System Configuration (AI model, poll interval, tenant) is read-only.