Very, the Verybear mascotverybear

How it works

From sign-up to synced, in six steps.

Verybear sits between the tools your people already use and the AI providers behind them. Here's what setting it up actually looks like.

1

Sign up & connect

Create your org, add the first admin, and connect the tools your people already use. No migration, no new app to install.

Connect your tools

Each person connects once. No API keys to copy around.

  • Copilot Connected
  • Claude Connected
  • Codex Connected
  • Cursor
  • OpenCode

$ verybear login

Open verybear.com/device and enter code BLUE-7FOX

2

Define flows

A flow is a task, named in plain language, with the model, effort, playbook and any helper agents that should run it. Write it once.

New flow

Task description

Take a ticket to a reviewable PR: plan, implement in a worktree, run the checks, write the PR body.

Model

opus

Effort

high

Playbook

Coordinator

Subagents

PlannerImplementerVerifier
3

Set guardrails

Approve the models you're comfortable with, give every team a budget, and block the request types that should never reach a model.

Approved models

SonnetOpusGPTGPT miniGeminiArch (local)

Budgets per team

Finance$2,310 / $4,000
Engineering$11,420 / $18,000
Support$3,875 / $5,000
Marketing$1,190 / $3,500
Operations$2,460 / $2,500

Blocked request types

  • Customer PII export
  • Payroll data
  • Source code to unapproved vendor
  • Legal advice to customers
4

Invite people

Add people by email, team and role, or turn on SSO and let your identity provider do it. No one is ever handed an API key.

Invite people

Add an email, pick a team and a role. That's it.

[email protected]

finance · Member

Pending

[email protected]

engineering · Member

Pending
[email protected]Member

SSO (SAML / OIDC) provisions people automatically on Business and above. No API keys are ever handed out. People sign in with their existing accounts.

No setup required

Or start in Learn first.

Skip steps 2 and 3 for now. People keep working exactly as they do today, and Verybear learns what to automate from what actually happens.

1

Observe

Everyone keeps working as usual. Verybear routes unknown tasks to a safe default and applies spend limits behind the scenes.

2

Auto-flow

After about 50 requests, Verybear proposes starter flows for each team, built from what people actually asked for.

3

Pilot

Pick one group, turn a flow on, and review results in 2 to 4 weeks before rolling it out further.

New flowAuto-flow proposal

Reconcile ledger to bank

Built from 54 similar requests in Finance. Suggested model: Claude, effort: Medium.

Feedback

It asks, then it improves.

Short, specific questions show up right after a flow runs, in the agent session, the desktop app, or on the web. Answers turn into suggested edits in the console.

In the agent session

Very: How did that go?

2 quick questions to improve Reconcile ledger to bank → verybear.co/c/8f3k

In the desktop app

Quick feedback

Reconcile ledger to bank

Did this save you time?

YesSomewhatNo

Was the output ready to use?

YesNeeded editsNo

The user experience

One prompt. Two very different views.

People just ask for what they need. Admins see exactly how it was handled, including the requests nobody had a flow for yet.

What everyone sees

Their normal tool. A plain prompt. Nothing new to learn.

I need to reconcile the Q3 ledger to the bank export
Running Finance · Reconciliation your way
summarize this ticket for engineering
We don't have a flow for that yet. Running General assistance safely

What the admin sees

Every request, with the flow, model and spend behind it.

  • Marcus Lee

    Reconcile ledger to bank · Sonnet

    $1

  • Sofia Alvarez

    Draft customer reply · GPT mini

    $0

  • Tom Okafor

    Review pull request · Sonnet

    $1

  • Dev Patel

    General assistance · GPT mini

    Flagged$0

  • Hannah Weiss

    Campaign brief · Gemini

    $2
6

Discovery & review loop

Every unhandled request and every pattern in the data becomes a proposal: a new flow to create, or an existing one to make cheaper or better. Approve it, and the change syncs to every connected desk.

Discovery inbox

7 to review
  • New flow

    Summarize ticket for engineering

    37 similar requests in the last 7 days hit the General assistance flow.

  • New flow

    Explain a Slack thread decision

    19 requests asked to summarize a thread and extract the decision.

  • New flow

    Draft a job description

    12 requests from Operations and Engineering.

  • Optimization$390 / mo

    Explain this codebase → move to Arch (local)

    Quality signal unchanged in a 2-week shadow run; 98% of prompts answered from local context.

Synced to 148 desks

Who it's for

Everyone gets something different. Nobody gets left out.

CFO

  • A spend number that matches the invoice
  • Per-team budgets that actually hold
  • Blocked request types, enforced everywhere

AI lead

  • One place to define every flow
  • New flows and savings proposed automatically
  • Nothing to retrain, nothing to re-explain

Everyone

  • The tools they already use, unchanged
  • No model picker, no API key, no training
  • The best way to do the task, by default

See it running on your own flows.

Start freeTalk to us