Admin Guide

Calibrate & tune Odly

Once the basics work — channels connected, agents replying, mail routing — this is how you make Odly sound and behave like your team. Calibration is admin-facing tuning: a handful of settings you adjust after setup, plus a clear line between what you edit, what Odly learns, and what stays a fixed default. Do the earlier setup steps in the Product Guide first; come here to refine.

Every path below lives under Settings. Prompts and the suggestion queue need moderator rights or higher (the Manage AI Prompts permission); Learning Trust, routing keys, auto-assign, providers and brand terms are org_admin-only. If you don't see a section, you likely don't have the permission.

Voice

1 · Reply voice & style

The one place to shape how Odly's drafts sound. Manage AI Prompts

Reply voice is an editable template, not something Odly learns. Open Settings → AI → Prompts and write the template named reply_style (up to 2000 characters). Use it to set tone, greeting and sign-off conventions, formality, and any house phrasing rules.

Additive, not a replacement. Your reply_style layers on top of Odly's built-in reply guardrails — it refines the voice, it doesn't switch the guardrails off.
Inactive by default. A fresh org has no reply_style, so drafts read in Odly's default voice until you write one. Nothing changes until you do.
Org-wide, with department overrides. Set one voice for the whole organization, then override it per department where the tone should differ. A department template wins over the org-wide one.

The same Settings → AI → Prompts area also holds the analysis prompts — spam, message analysis, lead qualification, and follow-ups. Those steer how Odly reads incoming mail, separately from how it writes replies.

Routing

2 · Routing keys, skills & auto-assign

Control how conversations reach agents. Admin

Routing keys (skills). Define org-level skill keys used for auto-assignment in Settings → Workspace → Routing Skills. Optionally let agents edit their own skills with the Allow agents to self-edit skills toggle on the same card.
Auto-assign mode. The Auto-assignment mode picker on the same card offers Always (the default), Match only (assign only when a skill matches), or Off.
Brand terms. Product and brand words (up to 50) are excluded from learned routing keywords — so the learning engine never turns your own brand name into a routing rule. There's no settings screen for this yet; it's the org-admin-only brand-terms API (PATCH /api/organizations/brand-terms).

Add brand terms early. Your company and product names show up in almost every message, so without this list the engine can mistake them for a routing signal.

Automation

3 · Auto-reply & lead qualification

Decide how much Odly does on its own. Admin

Auto-replycontrols whether the AI may draft and send automatically. It's set org-wide with per-department overrides — any department can turn auto-reply off. On Free, Odly only offers suggestions; automatic sending is a paid capability (Pro).

Lead qualification (Pro) tunes how Odly handles sales-shaped mail:

Required contact fields — which of name, email, phone, and company a lead must provide.
Qualification fields & categories — the questions and buckets used to score and sort leads.
Speed-to-Lead — a slow-lead threshold, your average lead value, and the recipients for the weekly digest.

Content

4 · Knowledge base & Confluence

The source material Odly draws answers from. Admin

Curate your knowledge-base sources so drafts cite the right material. Beyond answers you save from resolved conversations, you can connect Confluence Cloud as a live content source.

Provide the Confluence base URL, an email, an API token, and the space keys to include.
Optionally pick the specific pages and folders to include from the page tree, instead of whole spaces.
Odly syncs them into the knowledge base and keeps them current as your docs change — or press Sync now to refresh straight away.

5 · Vision & image analysis

Bring-your-own image analysis. Admin

In Settings → Integrations → AI Providers, the Vision card controls image analysis: switch it on or off, and on Bring your own keys pick the vision model. On managed AI, Odly uses an optimized vision model automatically.

Vision is wired to Telegram image attachments and knowledge-base scanned-PDF / image processing — not to every channel. Enable it when those inputs matter to you.

Learning & limits

6 · The learning engine

Rules that Odly proposes from what your agents actually do. Admin

Routing and spam rules aren't all hand-written — the engine proposes them by watching how your team works. Settings → AI → Engine Activity lists pending suggestions and recent auto-actions — moderators can act on the ones in the domains they manage. How much the engine may do on its own is the org_admin-only Settings → AI → Learning Trust mode: Manual (suggestions only), Balanced (proven rules auto-promote within limits), or Aggressive (rules promote more readily). Approved learned rules land in the Auto-learned group of Settings → Rules → Routing Rules. The consistency of your team's day-to-day actions is what tunes this the most.

For the full review-and-approve workflow, see the learning-engine section of the Product Guide.

7 · What you can't tune

A precise note so you don't go looking for knobs that aren't there.

Most similarity and confidence thresholds are fixed code-level defaults, notadmin-editable settings. There is no UI slider for them — treat them as Odly's internal defaults, not something to tune.

A threshold does live on some individual rule rows (a confidence_threshold, default 0.75, on contradiction-detection rules), but no settings screen exposes it — it's reachable only through the API or the database, not something admins tune day to day. Everything else in that family is a fixed default.