Policies only apply when your agent checks in with Troxy before acting. If a checkpoint has never been reached, its policies are not protecting anything yet.
Applies to everyone in the workspace, in every space.
Applies only to your space's members.
i
API Keys i
One key per computer or agent you connect to Troxy
Create a key, then run one command on the computer where your agent runs.
The engine works with Claude Code on Mac and Linux.
Name
Key
Last used
Expires
Status
Insights i
AI activity and decision trends
Last 30 days
Today
Last 7 days
Last 30 days
Last 90 days
Loading…
Usage and Savings i
How much further your Claude usage goes with Troxy
All agents
All agents
Last 30 days
Today
Last 7 days
Last 30 days
Last 90 days
troxytroxy
Chat
A private chat with the Troxy team. We usually reply within a few hours.
T
Troxy team
Real people answer here. We usually reply within a few hours.
Open
This conversation was marked resolved. Sending a new message will reopen it.
Quick answers
How do I connect Troxy to Claude Code?
Create a key on the API Keys page, which shows the install steps for your computer. Then run troxy init --key YOUR_KEY. Claude Code may ask you to sign in again, that is normal.
How do I update Troxy?
Run troxy update in your terminal, then type /reload-plugins in Claude Code.
How do I turn off model switching?
In Claude Code type /troxy off to stop Troxy choosing a model for you, so your own pick is used. Type /troxy on to turn it back on and /troxy status to check.
Where do I see what Troxy saved me?
On Usage and Savings. It is an estimate based on list prices, and it only counts from the moment Troxy started reporting.
What does Troxy keep about me?
Counts and model names, such as which model was used and how many tokens. See the privacy page for the details.
How do I remove Troxy?
Run troxy uninstall. You can delete your account in Settings.
Configure Troxy to watch and act for you automatically.
Your rules
⏱
No rules yet
Describe a task above and Troxy will watch it for you around the clock.
Settings i
Account and preferences
Account
Email address
Loading...
Profile
Used to fill in your name and phone when an agent needs them, so it doesn't have to ask you.
AI Policy Builder model
Always Claude Haiku, the fastest and cheapest tier. Policy conditions don't need a bigger model.
Powered by Claude
Claude Haiku 4.5
Current session
-
Onboarding & tour
Replay the setup checklist, or a quick walkthrough of Activity and Usage and Savings.
Plan
Free beta, no limits during beta period. More plans coming after beta.
Beta
Solo workspace
Built for one person. Want to invite teammates? Request team workspace mode and we'll set it up.
What Troxy does
Turn any of these off and Claude Code stops using it from your next answer. Needs troxy-cli 1.30.24 or newer: run troxy update, then /reload-plugins in Claude Code.
Loading…
My AI tools
The AI coding tools Troxy found on your machine. Run troxy tools to re-scan after installing something new.
Loading…
Notifications
Choose which emails Troxy sends you.
Notification email
Leave blank to use your account email
Email on block
Event blocked by a policy rule
Email on escalation
Event held for your approval
Email on notify
Event matched a Notify rule
Email on allow
Event approved, high volume, off by default
Weekly savings summary
Mondays, what Troxy saved you that week. Skipped if you did not use it
Near your usage limit
When your 5-hour or weekly Claude limit passes 90%, once per window
Escalation
Approval window
How long an escalated event waits before it auto-declines for safety
1 hour
5 minutes
1 hour
4 hours
8 hours
24 hours
3 days
No expiry
Danger zone
Clear all data
Deletes all events and reports. Your account and API keys stay active.
Delete account
Permanently deletes all API keys and activity, agents stop immediately
Delete workspace
Permanently deletes your workspace and its AI provider key. Only allowed once every other member has been removed - you can't take anyone else's access with it.
Workspace i
Loading...
Your workspace
Loading...
Join code
Share this with people you want to add. Regenerating invalidates the old code immediately.
Loading...
Domain restriction
Only allow people with emails from specific domains to join. One domain per line (e.g., @company.com).
Require manual approval for new joiners
Pending approvals
People who requested to join and are waiting for admin approval.
Invite by email
One email per line. Everyone on this list joins with the role picked below.
Workspace Admin
Workspace Admin
Workspace Owner
Members
MCP Connectors
Push Troxy's MCP config to every agent in your workspace. Zero install for users.
Connectors use each platform's admin API to register Troxy as an MCP server
and assign it to all workspace members instantly.
AI provider
Optional. Nothing else in Troxy needs this. Connecting one unlocks an extra recommendation for every member of this workspace that catches a turn that used an expensive model for a simple task. Set once here instead of every member pasting their own key. Only admins can view or change this.
Choose a provider:
Billing keys
Optional. Without this, your spend numbers already work. They're calculated from real token counts (token-math), just not pulled from your actual invoice. Connecting a billing key here replaces those estimates with your real bill, workspace-wide. This only ever applies to workspace totals and per-model breakdowns. It never attributes to a specific agent. Only admins can view or change these.
This needs an admin-level API key from the provider's own console, not a regular one - and not every account tier offers one. Anthropic's Admin API, for example, is unavailable on an Individual account; it requires a Team or Business workspace there. If a sync fails with "invalid or revoked" but you're confident the key is correct, check your provider account's plan tier before assuming the key itself is wrong.
Loading...
Space policies
Members
Add an existing workspace member to this space. Adding them here moves them out of any other space first - membership is one space at a time.
Choose a member
New local agent
For an agent on a machine you can open a terminal on. You will get a command to run there.
This is the name you'll see in your dashboard, and on every decision this agent makes.
Step 1 · Install Troxy (once per machine)
npm install -g troxy-cli
Needs Node.js 18+. On Mac or Linux, put sudo in front if you get a permission error (it asks for your computer's password). On Windows, run it in PowerShell or Command Prompt.
Step 2 · Connect this agent
No terminal access, or a one-off/cloud agent? Skip step 1 and run
instead.
New policy
How do you want to start?
What is this policy about?
Describe the policy you want
Troxy writes the rules; you review them before saving.
About-
Rules
No rules yet. Add one and either describe the condition or choose it from lists.
Describe one event and see what this policy would decide. Nothing is saved and no agent is affected.
This governs the email checkpoint on whichever agents below can send email - pick which ones.
This governs the action checkpoint on whichever agents below can delete, overwrite or destroy things, pick which ones.
This governs the login/signup checkpoint on whichever agents below can authenticate to sites - pick which ones.
This governs the model checkpoint: which model an agent runs on, and what that run costs. It only sees what the agent reports, so an agent that never checks in is not covered. It also does not yet apply to traffic routed through live model routing (troxy init) - that path only does automatic token savings (clearing/trimming/model swap), not policy enforcement.
This governs the secrets checkpoint: it scans agent output for API keys, credit cards, private keys and tokens before they are sent, written, or published. Pick which agents below are covered.
Choose a template
Each template comes with the action we recommend: block when there is no safe version of it, ask you when it depends, notify when you just want to know. Change any of them after adding.
Connect Claude Code
Enter the code Claude Code shows you.
⚠️
Delete your account?
This will permanently delete all your activity history and API keys. Your agents will stop working immediately. This action cannot be undone.
⚠️
Delete this workspace?
This will permanently delete the workspace and its AI provider key. This action cannot be undone.
Clear all data?
This will permanently delete all decisions and reports. Your account and API keys will stay active. This cannot be undone.
This only clears your own activity. Usage and Savings shows every active member of your workspace, so it can still show non-zero numbers afterward if teammates have their own usage.
Edit rule
New rule
Autopilot can watch and notify you. Automatic purchasing is coming soon.