Frequently asked questions

Trellis runs on Claude, made by Anthropic — the same AI family trusted by security-conscious enterprises in healthcare and finance. Anthropic’s most capable models draft feedback reports and coaching suggestions, and a faster Claude model cleans up observation notes and powers Elli, our built-in assistant.

OpenAI models provide automatic backup, plus voice-note transcription and search indexing — so Trellis keeps working even if one provider has an outage. We keep models current as new versions ship.

No. We use Anthropic and OpenAI through their commercial APIs, and under those terms customer data is not used to train their models.

On top of that, before any text leaves Trellis, teacher names have already been replaced with anonymous codes.

Only what it needs: the observer’s notes and the teacher’s recent feedback history. Before anything is sent, Trellis swaps the teacher’s name and email for an anonymous code — something like “Teacher-4f8a2c” — generated with salted one-way cryptographic hashing, so it can’t be reversed into a name. The real name is restored only once the text is back inside Trellis.

There’s also a built-in tripwire: if a teacher’s name would slip through, the request is blocked outright. This anonymization runs on every AI path — note enhancement, feedback drafting, coaching, Elli, and search indexing.

Trellis is built so nothing reaches a teacher unchecked. The AI works only from what your observer actually wrote, plus the teacher’s documented history — and nothing is delivered until an administrator has reviewed it, edited whatever they want, and explicitly chosen to send it.

A person is always the author of record.

No — and this is core to what Trellis is. Trellis is a growth and feedback platform, not an automated evaluator. The AI drafts; your leaders decide.

It never assigns ratings on its own, and nothing is pushed automatically into personnel files or HR systems.

Every new feedback report draws on the teacher’s recent observations and prior feedback, so feedback builds instead of starting from zero — that’s how Trellis can say “this is the third visit in a row where students owned the discussion.”

And Elli can search across your school’s whole feedback archive to answer questions and spot trends.

Elli is the AI assistant inside Trellis. Ask it something like “What trends are we seeing in 4th-grade math?” and it answers using only your school’s own observations and feedback — the school boundary is enforced on every query, so it cannot see other schools’ data.

On the District plan, district leaders can ask across all of their schools.

You do — unambiguously. Your admins can export the school’s complete dataset at any time, as spreadsheets (CSV) or JSON, right from the product.

We never sell data, never use it for advertising, and never use it to train AI models.

Trellis is designed to support your FERPA obligations, and we’ll sign a data processing agreement that puts our responsibilities in writing. Worth knowing as you compare vendors: there is no official “FERPA certification” — no government body certifies vendors.

Trellis primarily holds staff observation and feedback data — not student education records. There are no student accounts and no student roster. And deletion rights are built into the product: removing a user cascades through their observations, feedback, search index entries, and chat history.

No student accounts, no student roster, no student records — Trellis is a tool for adults, about adult practice.

Students may be mentioned in passing in an observer’s narrative notes; those notes get the same encryption and access controls as everything else, and many schools keep student references to initials as a matter of practice.

The recording is uploaded to private storage, transcribed to text, and the audio file is deleted immediately. Trellis keeps the written transcript — not the recording.

You export everything, then we delete everything. Admins can pull the full dataset self-serve, individual users can delete their own accounts, and the workspace itself can be deleted — and we keep a record that the deletion happened.

Deletion from live systems is immediate; copies in our infrastructure providers’ managed backups expire on their standard schedules.

Inside your school, access is role-based: teachers see the feedback delivered to them, school admins see their school, and district admins see their schools.

On our side, production access is restricted to our engineering team for support and incident response. Even our server logs are built for privacy — identifiers are hashed, and note content stays out of logs entirely.

Every layer Trellis runs on is SOC 2 Type II audited — our hosting, database, identity, payments, and AI vendors all maintain independent third-party security audits. Trellis itself is an early-stage company and hasn’t completed its own SOC 2 audit yet; it’s on our roadmap.

In the meantime, we’re happy to share our security documentation, sign a data processing agreement, and complete your district’s security questionnaire.

Yes — encrypted in transit with TLS and at rest with AES-256.

Payment details never touch our servers at all: checkout is handled end-to-end by Stripe (PCI DSS Level 1), the same processor behind much of the internet’s commerce.

Every record in Trellis is tagged to your school’s workspace, and every single query is filtered by that boundary — a user in one school cannot reach another school’s data.

The AI search index is scoped the same way.

Sign-in runs on Clerk, a dedicated identity platform that is itself SOC 2 Type II audited — Trellis never stores passwords. Multi-factor authentication is supported, Google sign-in works out of the box, and SAML single sign-on is available on District agreements.

A few of the safeguards working behind the scenes:

• Every request is authenticated and its payload validated before it touches data.
• Rate limiting on sensitive endpoints blunts abuse and brute-force attempts.
• Strict browser protections: Content-Security-Policy plus anti-clickjacking and anti-MIME-sniffing headers.
• Uploaded files live in private buckets, served only through short-lived signed links — never public URLs.
• Secrets and server keys exist only server-side, and logs are scrubbed of identifiers and note content.

On enterprise cloud infrastructure: the application runs on Vercel, and the database is managed PostgreSQL on AWS via Supabase. The database is backed up automatically, with point-in-time recovery available.

There’s no on-premise install and no servers for your IT team to maintain.

The platforms Trellis runs on deliver better than 99.9% availability historically, and the architecture scales automatically — there’s no single server to fall over.

For District agreements, uptime commitments are part of the contract conversation.

Nothing. Trellis is entirely browser-based — laptops, tablets, phones.

In districts that filter web traffic, the only ask is allowlisting the Trellis domain.

Five steps, and you control every one of them:

1. Capture — the observer types or dictates notes during or after the walkthrough, and can attach artifacts like photos of board work or handouts.
2. Enhance — in seconds, the AI turns raw notes into a clear, structured, framework-aligned feedback draft.
3. Review — the administrator reads the draft, edits anything, and decides what to deliver. Nothing is shared until they do.
4. Deliver — the teacher sees the feedback in their own portal, acknowledges it, and can reflect and respond.
5. Compound — the next feedback report builds on this history, so growth is visible across the year.

Yes. Danielson, Marzano, CSTPs, UDL, your state rubric, your local walkthrough form — your framework is configured at the school level, and every piece of feedback aligns to it.

You can also upload your own guidelines and look-fors for the AI to honor.

There’s no live SIS or LMS sync today — and that’s partly by design: Trellis doesn’t need your student data, which keeps the privacy footprint small.

Setup is light: invite staff by email, and move data in or out as CSV. If a specific integration matters to you, tell us — that’s exactly the roadmap input we want.

Same day. Create the workspace, invite staff by email, start observing.

There’s a 14-day free trial — no credit card to start — and most schools run a real observation cycle inside the trial.

The Site plan is $249 a year — unlimited teachers, unlimited observations and feedback, and Elli included.

District pricing is custom, based on size and needs, and adds district-wide analytics, cross-school visibility, and priority service.

Yes — District agreements run on purchase orders and invoicing with standard NET-30 terms.

Self-serve site plans are card-based through Stripe, with monthly or annual billing.

Your data is safe — the workspace pauses until you subscribe, and billing settings stay reachable so an admin can upgrade in two clicks.