Three minds.
One answer.

LookinGlass runs three top AI models — Claude, GPT-4.1, and Gemini — side-by-side. Compare what they say. Synthesize what they share. Stop guessing which model is right.

3 frontier models
1ms tie-break
second opinions
See it run

Parallel mode, side by side.

One question. Two answers. Streaming in real time so you can watch them agree, diverge, and disagree on the things that matter.

What's the best ratio for a perfect espresso shot?
Claude
GPT-4.1
Parallel Streaming · 2 providers
Three ways to ask

Standard. Parallel. Bash.

Pick a mode per question. Single-model speed for everyday work. Side-by-side comparison when stakes go up. Synthesis when you need the best answer money can buy.

Default

Standard

One model with tools — web search, document/image, Office 365, calendars. The mode you'll use 80% of the time. Fast, conversational, low-cost.

Compare

Parallel

Two models answer in parallel. See where they agree and where they don't — disagreement is signal. Use it for decisions, drafts, and "is this right?" gut-checks.

Synthesis

Bash

Three models reply. A fourth judges. You get a consensus answer with the strongest argument from each model. Slower, more expensive, the most reliable.

The lineup

Four frontier minds. One workspace.

Three top-tier models running side-by-side, plus a premium creative-writing model in private beta. We use the API tiers that contractually exclude training — your data stays yours.

Claude

Anthropic · primary speaker

Long-context reasoning, code, structured writing. The default in Standard mode.

GPT-4.1

OpenAI · counterpoint

Fast, conversational, often picks up things Claude misses. Strong on creative.

Gemini

Google · wide-context

Cross-domain knowledge, long-doc summarization, factual triangulation.

Beta

Fable 5

Anthropic · premium narrative

Long-form creative writing, stylistic prose, narrative drafting. Opt-in per user.

What you get

Built for people who actually use AI.

Not a chatbot toy. A serious workspace for knowledge work, with the features that stop showing up in every-other-AI's roadmap.

M

Memory that compounds

LookinGlass remembers what you tell it across every chat. Auto-extracts facts. You pin the ones that matter. Every new conversation starts already knowing you.

P

Projects + system prompts

Group chats by client, topic, or domain. Each project has its own system prompt and reference docs. Drop a PDF in; every chat in that project sees it.

📁

Office 365 native

Word, Excel, PowerPoint, PDF — read, summarized, transformed. Backend uses LibreOffice headless so the layout doesn't get mangled. No screenshots required.

🔔

Push when ready

Ask a Bash-mode question. Close the app. Get pinged on your lock screen when all three models finish. Nobody else does this.

🎙

Voice in. Voice out.

Dictate questions in any language. Voice mode reads answers back so you can multitask. Five personality presets pick the tone.

🔒

Your data, your data

We don't train on your conversations. None of our model providers do either — we use API tiers that contractually exclude training. Delete it all in one tap.

For teams

Built for the regulated middle.

Single sign-on, audit logs, IP allowlist, account deletion — at the Team tier. Things Claude Team and ChatGPT Team make you upgrade to Enterprise for.

Workspaces

Multi-tenant from day one. Each workspace gets its own SSO, brand color, welcome message, and default mode. Users can belong to many.

SSO via any IdP

Zitadel-backed OIDC means anything that speaks OpenID Connect works — Okta, Microsoft, Google, custom. Not just the two SSOs everyone else supports.

Audit log + IP allowlist

Every action recorded server-side. Lock workspaces to corp-net IPs. Required by your security team, included at the Team tier.

Delete + export

Users can delete their account in two taps. Export all their data to JSON. GDPR-friendly without paying the GDPR-friendly premium.

Team · most popular

Workspace

$29/seat/month
Billed annually · 5-seat minimum
  • All 3 modes (Standard, Parallel, Bash)
  • Memory + Projects + Personalities
  • Office 365 + PDF + image ingestion
  • Workspaces, SSO, audit log, IP allowlist
  • Push notifications + iPad split view
  • Account delete + data export
  • iOS, Android, web
Request access →
Head to head

Everything they do — plus the things they don't.

Feature parity with both ChatGPT Team and Claude Team — and the one thing neither does: run them side-by-side.

Feature LookinGlass ChatGPT Team Claude Team
Multiple models in one chatOnly us
Synthesis across models Only us
Memory across chats
Projects + system prompts
Office 365 native ingestion
Workspaces + multi-tenant
SSO at Team tier Any OIDC IdPGoogle/MSGoogle
Audit log at Team tier Enterprise onlyEnterprise only
IP allowlist at Team tier Enterprise onlyEnterprise only
Push when reply is ready Only us
Account deletion + export
Questions

Short answers, no hand-waving.

If yours isn't here, get in touch.

Which models does LookinGlass actually run?
Three top-tier frontier models: Claude (Anthropic), GPT-4.1 (OpenAI), and Gemini (Google). Models update as new versions ship. Fable 5 — a premium creative-writing model from Anthropic — is in private beta and can be enabled per user on request.
Does my data train any model?
No. We use the API tiers from each underlying provider that contractually exclude your inputs and outputs from being used as training data. All three frontier providers offer this tier and we use it for every request. Receipts available on request for your security team.
What's "Bash mode," and why is it slower?
Bash sends your question to all three models in parallel. When they all finish, a fourth pass synthesizes their answers into a single response — citing where they agreed and where they disagreed. Slower because it waits for the slowest model. Better answer because three minds beat one when stakes are high.
Why is this better than just using each AI separately?
Because copying a question between tabs is friction, and humans rarely do it. LookinGlass makes "ask both" the default — and shows them side-by-side so disagreement is visible. Disagreement is signal: if Claude and GPT give different answers, the question is harder than you thought.
Do you support Office 365 / Outlook / Slack?
Yes — Office and PDF are native (LibreOffice headless on our backend, not screenshots). Outlook calendar and email read access via your workspace SSO. Slack send-message tool live. More tools shipping monthly.
Is there an iPad app?
Yes. The iOS app is native on both iPhone and iPad — on iPad it runs as a proper split view with the chat list permanently on the left. Designed for the Liquid Glass aesthetic; it really pops at iPad size.
How does pricing work? Can I try it first?
Private beta right now — invite-only. Once we open up, Workspace is $29/seat/month billed annually with a 5-seat minimum. There'll be a 14-day trial. No free tier forever; we pay the model providers per token, so quality stays the priority.
Can I self-host LookinGlass?
Not today. The underlying provider APIs are SaaS, so even a self-hosted LookinGlass would call out to them. We're exploring an on-prem tier for regulated industries — get in touch if that's the only way you can buy.

See what three minds
think of your question.

Private beta. Request access — we'll usually approve within a day.

Request access →