Connect your stack once
Stripe, GitHub, Jira, Zendesk, Amplitude, Slack, and dozens more.
Atmosphor presents Vernais
Know what to build next.
Artificial Organizational Intelligence for better decisions.
The problem
You have every tool. You still can’t answer “why.”
Eight dashboards, all open, none agreeing. The answer is in there somewhere, scattered across tabs that were never built to talk to each other.
How it works
One brain. Every tool. One question at a time.
Ask in plain English
One chat box, plain English, no query language.
Act and measure it
The answer becomes a plan, the plan becomes shipped work, the work becomes a tracked outcome.
The Company Brain
Every tool you use, resolved into one graph you can query.
40 SaaS sources flow in through Airbyte-grade connectors. Vernais cleans them, deduplicates them, and links them into three kinds of nodes:
One clean, deduped record.
A Jira issue. A Stripe charge. A Slack message. The atomic unit of evidence.
A resolved real-world thing.
Acme Corp. onboarding-service. v2.2.1. Mentioned in five tools, it appears once, and becomes the glue that connects them.
An ML-discovered cluster.
Organized umbrella → subtopic. Nobody typed the categories. The math found them.
Learn more: The Company Brain
The Investigator
Ask “why,” and get a cited answer, not a dashboard.
Behind the chat box is an 11-stage reasoning engine. Ask a “why” question and watch it work, live.
It decomposes your question into sub-questions. Scopes to the right topics. Recalls the connected evidence with Personalized PageRank. Weighs competing explanations through a deterministic node-jury. Debates them PM-versus-Engineer. Then it writes a cited root-cause Case file with measured impact. The moment it finds the cause, it derives a North-Star Tracking metric to track the fix.
You see the reasoning stream, stage by stage. You see the citations. You can check the receipt on every number.
Learn more: The InvestigatorInitiatives
From finding to shipped: the five-phase spine.
Identify → Validate → Decide → Act → Measure.
A finding isn’t the end. It’s the start of an Initiative: a structured plan with a Jira-style ID (IN-42) that moves through five phases. Every hypothesis, solution, and task carries a human-readable ID you can point to.
The workspace
The surfaces that replace your tools.
The work doesn’t happen in a silo anymore. Vernais brings the connected workspace inside the brain, and everything routes back to the initiative. Tag any page, message, or transcript to an initiative. Vernais correlates it, classifies it, and routes it to the right phase, automatically.
Replaces Notion
Codex
Notion-grade docs. 20+ block types, inline databases, subpages, comments, and live initiative-dashboard blocks.
Replaces Slack
Wire
Your workspace as a server. Channels, DMs, threads, reactions, mentions, opt-in E2EE for direct messages.
Replaces Google Meet
Beam
One-click video rooms with in-browser transcription. The meeting ends; the transcript becomes evidence.
Replaces Jira
Forge
A Kanban board and Gantt plan with three flow state machines and an AI task decomposer.
Automatic routing
Evidence travels. Automatically.
Nothing you write is stranded. A doc, a message, a meeting; each one becomes graded, classified evidence on the initiative it belongs to.
A Page
A scored, classified initiative signal
Lands inValidate (Correlations), where a solution surfaces in Decide
A Wire message
A correlated signal with author captured
Lands inValidate, where a hypothesis match bumps its evidence
A recorded meeting
A speaker-labeled transcript, correlated and classified
Lands inValidate + a restricted transcript page in Codex
A fix proposed in a chat or doc
A signal-born Solution linked to its source
Lands inDecide (a new card for your review)
A completed investigation
A full initiative: root cause, hypotheses, signals, metric
Lands inIdentify
Trust & provenance
Every number has a receipt.
This is the difference between an AI that guesses and one you can trust.
Vernais stores only grounded facts: each with its source span, its signal ID, its confidence, and which check it passed. Causality is never baked in. “X caused Y” is computed fresh at query time, from the timeline, and NLI-checked before you see it.
Facts stored once, causality derived.
The graph holds what’s grounded; the “why” is computed on demand.
A closed-menu grounding lock.
A derived metric can only pick from real events the field index found. The model cannot invent one.
Same question, same answer.
A seed-locked jury, cached hypotheses keyed to your data, reproducible scores.
It says “I don’t know.”
No evidence returns an honest answer, not a fabricated one.
Security
Enterprise-grade from the first login.
Complete workspace isolation
Each workspace is its own pair of databases. Membership is re-checked on every call.
Encrypted where it matters
Argon2id passwords, opaque hashed sessions in HttpOnly cookies, AES-256-GCM on credentials and chat.
Google Sign-In, hardened
OAuth2 with PKCE and OIDC.
Roles and audit
A 5-role permission model, a full per-workspace activity trail, and a separate immutable admin audit log.
Reporting
The living report.
The initiative is a document that grows: from problem statement, to locked metric, to shipped result.
Dashboard
One-glance overview: KPI strip, throughput, recent activity, decision-ready initiatives, sync health.
Launches
Every tracked metric in one outcomes view: improving, regressing, awaiting, broken down by owner, impact, and product.
Reports
Consulting-grade views synthesized from an initiative or framed as a new question.