AI governance for software teams · built on Google Cloud

Every role gets the right AI. The company keeps control.

RoleSpace gives product managers, designers and developers the AI help their job needs, from checked answers and safe edits to a full coding workspace. One policy decides models, repositories, budgets and review for everyone.

See how it works

Runs in your own Google Cloud project
Google sign-in, no VPN
Every decision explained and logged

RoleSpace · FirstRound workspace

Models you can switch on

Gemini
Kimi K2
Qwen3 Coder
DeepSeek
Claude
all on Vertex AI

of developers spend more than 30 minutes a day searching for answers.¹

faster on a coding task with an AI pair programmer, in a controlled study.²

RoleSpace citations verified against the lines the assistant read.³

model cost per answered question, measured on RoleSpace.³

¹ Stack Overflow Developer Survey 2024 · ² GitHub research, 95 developers, one task · ³ RoleSpace evaluation on the running product, 17 Oct 2026: 10/10 questions, 10/10 attacks stopped, 12 s median answer

How it works

Four levels. One policy.

Each person gets a level from their team. The level decides their AI tools, the code they may change, the models they may call and what they may spend. Pick a person to see exactly what they can reach.

Ask

Checked answers

Questions about the code; every citation is checked. No writes.

PMs, analysts, support

Suggest

Safe edits as pull requests

Small changes inside an allow-list, reviewed by a human.

Designers, writers, QA

Build

A private workspace

VS Code and Gemini CLI with only their repositories.

Developers

Govern

Control and evidence

Budgets, approvals, usage, a kill switch and the audit log.

Platform and security admins

Try it

Who can do what? Ask RoleSpace.

Pick a person and something they want to do. RoleSpace runs the same checks as production, in order, and explains the decision in plain words.

1 · Who

2 · Wants to

Policy checkrole.yaml · 7f1350e

Models & repositories

The right models and code for each person.

Switch on Gemini, Kimi K2, Qwen and more from Vertex AI, and decide who may work on which repository. Each person's key enforces the models; workspaces clone only their repositories; every change is logged.

Models by levelclick to switch on or off

What each key allows now

Collaboration

Changes from people without GitHub accounts.

A designer asks for a change in plain words. It reaches GitHub as a normal pull request in their name, with a CI check and a code owner's review. Nobody hands out repository access.

1Ask in plain words
2Review the diff
3Pull request opened
4Checked and reviewed
5Merged

Usage & cost

See every token. Cap every dollar.

Every model call goes through one gateway on the caller's own key, so tokens and cost are counted per person, model and project, and the caps hold even if someone bypasses the app.

Usagelive · sample data

Tokens in0

Tokens out0

Cost$0.00

Blocked0

Four caps, smallest first

$1–20 a montheach person's key
The gateway refuses their next call.
30–120 calls a minuteeach person's key
A runaway agent loop is slowed down.
One total for all AIeveryone together
The gateway refuses every call past it.
A project-wide hard stopthe whole cloud project
Once armed, billing is cut before the limit, so nothing can run up a bill.

Return on investment

What RoleSpace gives back.

Every question a PM can answer alone, and every small change a designer can make safely, is developer time handed back. Set the numbers for your company.

Your companydrag to change

Developer hours handed back a year0

Value of that time$0at your hourly cost

RoleSpace running cost$0

Return on running cost0×

Value vs running cost, per yearlive

Developer time handed back while you've been on this page, at your numbers$0.00

How we calculate this

Hours handed back = people using Ask and Suggest × questions a week × 48 weeks × developer minutes per question, plus small changes a month × 12 × developer minutes per change. Running cost = the app, gateway, database and sign-in at list price (about $1,300 a year) plus model calls at about 1¢ a question and 3¢ a change, measured on Gemini Flash. The live counter spreads the yearly value over 48 working weeks of 40 hours. Developers' own coding speed-up isn't counted.

Enterprise integration

Fits the way your teams already work.

RoleSpace runs in your own Google Cloud project and plugs into the sign-in, repositories and review process you already have. A pilot takes days, not a migration.

Security

Built for the security review.

Google sign-in at the edge

Identity-Aware Proxy; no VPN, no public IPs on workspaces.

A key per person

Own models, budget and limits; it can call models only.

Allow-lists in code

Checked when proposed, when the PR opens and again in CI.

Tamper-evident audit

Hash-chained events; evidence mapped to SOC 2 and ISO 27001.

Least privilege

Only the gateway may call models; one grant per secret.

Prompt-injection aware

Instructions hidden in files are data; unread citations are struck out.

Why RoleSpace

How RoleSpace compares.

Coding assistants, AI gateways and chat assistants are each good at their own job. RoleSpace is the layer none of them provides: one policy that ties AI to each person's role, from a PM's question to a developer's workspace.

GitHub Copilot

AI pair programmer, licensed per developer seat.

RoleSpace adds role levels for PMs and designers, checked citations and allow-listed edits as reviewed PRs.

Cursor

AI code editor for developers, per seat.

RoleSpace governs everyone around the editor: repositories, models, budgets and review per person.

Windsurf &

Cody

AI coding assistants and code search for developers.

RoleSpace explains every decision with the rule that decided, and keeps a hash-chained record.

ChatGPT Enterprise

A general chat assistant for every employee.

RoleSpace answers about your code with checked lines, and turns a request into a reviewed change.

{ }LiteLLM · Portkey

AI gateways: keys, budgets and routing.

RoleSpace builds on a gateway and adds people, repositories, file paths and review on top.

	Copilot	Cursor	ChatGPT Ent.	{ }AI gateways	RoleSpace
Built for PMs and designers, not only developers	Seat each	–	General chat	–	✓
Answers cite exact lines, and each citation is checked	–	–	–	–	✓
Non-developers' edits arrive as reviewed PRs inside an allow-list	Partly	–	–	–	✓
Models, budget and rate limits per person	Partly	Partly	Partly	✓	✓
Repository and file-path access per person	Partly	Partly	–	–	✓
Every decision explained, with the rule that decided	–	–	–	–	✓
Runs in your own cloud project	–	–	–	✓	✓
Pay for use, no seat for occasional users	–	–	–	✓	✓

From public product information, October 2026; products change often. Names and logos belong to their owners and are used only to identify the products. RoleSpace builds on an AI gateway (LiteLLM) rather than replacing one, and developers can keep their favourite tools: any tool that accepts a custom model endpoint can run on a person's RoleSpace key.

FAQ

Questions buyers ask.

Does our code leave our cloud?

RoleSpace runs in your own Google Cloud project. Your repositories are read there, and models are called through Vertex AI in the same project. A model from outside Google Cloud is only used if an admin switches it on with your own key.

Do designers and PMs need GitHub accounts?

No. Their changes are opened as pull requests on their behalf, in their name, and they follow checks and reviews inside RoleSpace. Code owners review in GitHub as usual.

How do we stop runaway AI spend?

Every call uses the person's own key, with a monthly budget and per-minute limits. A total cap covers all AI together, and a project-wide hard stop can cut billing before a set amount.

How do we add people?

An admin adds a person by work email and picks their level, projects, models and budget. RoleSpace creates their key and logs it. Directory sync with Entra ID or Okta is on the roadmap.

Bring AI to every role, safely.

Read the code