AgentWorks
GitHub Book a call

Product

Overview Goals Crews Code Relays Soon Auto-improve Connectors MCP gateway Soon

Use cases

All premade agents Sales & GTM Shopify stores Customer support Finance & billing Marketing & growth Engineering teams

Enterprise

Overview Release quality Incident response Security testing Cloud cost Growth analytics Code Expert Crews MCP gateway Soon Pricing Docs GitHub Book a call

Code

Every employee wants coding agents. Bring them to one place.

Today each person runs Claude Code, Codex or Cursor on their own laptop and their own plan, and the company can't see what it costs, who uses how much, or what the agents actually produced. With Code, everyone codes on your AgentWorks server, on the plans you already pay for, and you review cost, usage and output in one place. People can reach their Code from Slack or WhatsApp when they're away from the desk.

Book a call What you see

Coding accountsOn your server

Claude Max · AvaShared with 4 people and 2 Crews · 38% of weekly limit usedShared
Codex · team keyServer account · admins choose who can use itServer
Cursor · RioOnly Rio's own CodesPrivate

Usage this weekBy person and Code

R
Rio · checkout-serviceClaude Max · Ava's account
1.2M tokens
S
Sage · data-pipelineCodex · team key
640K tokens
O
Release notes CrewClaude Max · Ava's account
210K tokens
A
Ava · mobile-appClaude Max · own account
880K tokens

Illustrative example.

What the company sees

One place to review costs. And usage and output, for every person.

Work in a Code is private from colleagues and reviewable by the company. Everyone sees that stated plainly in the product.

  • Cost

    Cost and tokens per person, per Code and per provider account, so AI spend is a number, not a guess.

  • Usage

    A live usage and quota check for every account. Account owners see who used their account and where. Admins see everything.

  • Actual output

    Admins and named Code reviewers can open any Code's chats, files and costs, read-only, with every view audited. AI assistants can do the same over MCP, for example to summarise a team's week.

One place for every coding plan

Share the plans you already pay for. Instead of a seat for every person.

Connect each account once. Decide who can use it. See where every token went.

  1. Connect once

    Add Claude Code, Codex, Cursor or Muse by browser login, or any of them by API key. Credentials are stored encrypted on the server, and nobody sees them.

  2. Share without the login

    Keep an account private, or share it with chosen people, workflows or Crews. Admins decide who can use the server's own accounts and set the default models.

  3. Track it centrally

    Cost and tokens per account, per person and per Code, workflow or Crew, plus a live check of each plan's own usage limit.

  4. Stay accountable

    Runs on a shared account act as, and bill to, its owner. Owners see who used their account and where. Admins see everything.

On the go

Every Code can connect to Slack and WhatsApp. Keep working from your phone.

Connect a Code to a Slack direct message or a one-to-one WhatsApp chat. Ask for a fix, check a test run or get a summary without opening a laptop. The same Code, the same files, the same cost tracking.

WhatsAppcheckout-service

Did the retry change pass the tests?

Yes All 48 tests pass after the retry change. Want a short summary for Rio?

Yes, send it to me here.

checkout-servicePrivate · Claude Code

Add retries to the payment webhook handler, then run the tests.

Ran npm test in the sandbox · 2 failures, fixed, re-ran

Done Retries with backoff on 5xx and timeouts, capped at 3. All 48 tests pass. The change is in webhook.ts and its test file.

Who has accessOnly the owner shares

A
AvaOwner
Co-owner
R
RioReviews the changes
Editor
S
SageFollows along
Viewer
C
Code reviewerRead-only, every view recorded
Audited

A private workspace for everyone

Each person gets a Code. Their files, their agent.

  • The coding agent you like

    Claude Code, Codex, Cursor, Muse or Pi, on a shared account or your own.

  • Sandboxed commands

    The agent runs every command for you inside a Linux Landlock sandbox with its own private /tmp. Its own files are protected. There's no open shell on the server.

  • Private by default

    A new Code is yours alone. Skills you add stay in it.

  • Share with the right people

    Add someone as viewer, editor or co-owner. Only the owner shares, and removing someone takes effect at once.

  • Uses your Crews and workflows

    Your agent can ask a Crew or run a workflow. Nothing can call into a Code.

  • Cost you can see

    Tokens and cost tracked per Code, per person and per AI account.

Connections

Connect your own tools. They stay yours.

  • MCP servers

    Use the servers your admin selects for the Code, plus your own remote MCP servers. Your logins and API keys are stored encrypted, switched on per Code, and never usable by anyone else. Calls to private or internal network addresses are blocked.

  • Google, GitHub and more

    Add a known MCP server, such as Google Workspace or GitHub, as your own and sign in with your account. Personal secrets you add reach only your Code's chats.

Claude Code Codex Cursor Muse Pi

For teams

Private to people. Open to review.

A Code is private, not hidden. The people responsible for the server can review it, and the Code says so up front.

  • Admins see the whole picture

    Every account, every person's usage, and which Codes, workflows and Crews it went to.

  • Same accounts for Crews

    The plans you bring also power your Crews, in chat, Slack, WhatsApp or the live terminal. For Enterprise, sign-in goes through your SSO, with roles and per-workflow access.

  • Owners see their account

    Whoever shares an account sees who used it and where, and can stop sharing at any time.

  • Code reviewers

    Admins, and people an admin names as Code reviewers, can read any Code's chats, files and costs. Read-only, and every view goes in the audit log. AI assistants can do the same review over the AgentWorks MCP connection with a dedicated code:review token.

FAQ

Code, answered.

Which coding plans can we bring?

Claude Code, Codex, Cursor, Muse and Pi. Connect an account by browser login (Claude Code, Codex, Cursor, Muse) or by API key. It's stored encrypted on the server, and nobody sees the credential.

Who pays when someone uses a shared account?

The account's owner. Runs on a shared account act as that owner and use their plan, and the owner can see who used it and where.

Can we see how much of a plan is left?

Yes. Each account has a live usage check that reads the coding CLI's own usage or status, so you see the limit the vendor reports.

Is there a terminal?

No. There is no standalone terminal or shell. You ask the agent, and it runs the commands for you, each one in its own sandbox.

Who can see my Code?

You, and the people you share it with. Admins and anyone an admin names as a Code reviewer can also read its chats, files and costs, read-only, and every view is recorded. The Code itself tells you this.

How is it isolated?

Every command the agent runs is sandboxed with Linux Landlock and gets its own private /tmp, and the agent's own files are protected. The sandbox covers commands. Some coding agents' built-in read tools aren't sandboxed yet, so treat a Code as private to people, not sealed off from the server.

Can other workflows or Crews use my Code?

No. A Code can call your Crews and workflows, but nothing can call into a Code.

Bring your team's coding plans in. See usage in week one.

We set up AgentWorks with your team, connect the coding accounts you already have and open the first Codes together.

Book a call All products