They watch. We stop.

Every agent runs a tab. Hit the cap, the tab closes.

A hard stop for agent fleets, not another gateway. Run Claude Code, Codex, Grok or Kimi through tab, on your own subscriptions or your provider keys. Self-host the proxy and your prompts and keys go straight to the provider.

Start a flock Read how it works

One-line proxy swap. No SDK. Solo free (3 named Agents).

Example tab

tab_02c9 · my-project

OPEN

Spent

$3,412.60

of $5,000.00 hard cap

gpt-4.1 · 12,400 tok$0.31

claude-sonnet · 8,102 tok$0.19

embed-3 · 44,000 tok$0.04

reserve · pending$0.42

Closes at cap.Close tab

Works with your agents

13 coding agents run on a tab.

All 13 bill on a tab. Agent Guard checks tool calls in 10 of them.

With Agent Guard

Claude Code
Codex
Gemini CLI
Antigravity
Grok
Kimi Code
OpenCode
Goose
Cline
Pi

On a tab, without Agent Guard yet

OpenClawAgent Guard coming soon
AiderBilling only
Copilot CLIBilling only

Billing tested on tab 0.1.40 and Agent Guard on tab 0.1.43, 26 September 2026. Every lane: own key, flock key, subscription and Agent Guard.

The problem

A loop that retries forever spends forever. Alerts arrive after the money.

Provider budgets are monthly and shared. Dashboards report, they do not stop. By the time the alert fires, the invoice is already written.

Without FlockTab · one agent · one night

02:14 retry 1 · tool_call timeout$0.41

02:14 retry 2 · tool_call timeout$0.41

02:15 retry 3 · tool_call timeout$0.41

… 6,211 more retries

07:52 alert: budget at 80%$2,545.73

08:31 someone reads it$2,819.06

With FlockTab · cap $50.00

02:31 reserve $0.41 · blocked · capCLOSED at $50.00

In their own words

What people building with agents keep running into.

Surprise bills and overnight loops

Patrick T.

From X · Sep 2026

His app finally started getting some serious traffic, and surprise he woke up to a huge bill from Anthropic for his app’s Sonnet and Haiku usage.

Name withheld

From X · Aug 2026

if the state is still the same, another retry isn't persistence. it's just another bill.

FlockTab

A hard cap on every Agent's tab, reserved before each call, and a velocity limit that stops a loop.

How the cap works

No hard cap, no kill switch

Danilo O.

From X · Jun 2026

Gemini is currently charging us $1K per hour due a bug with the explicit cache feature, and I am unable to delete the cache from my end.

Elie S.

From X · May 2026

My credit card limit was the only way to break the infinite loop.

Cagatay U.

From X · May 2026

Default-on spend caps for Generative Language API.

FlockTab

Caps are enforced before the provider is called, not alerted on after. Close all tabs is one switch.

How blocking works

Subscription limits that shrink

Name withheld

From X · Mar 2026

use of Claude Code between 5am-11am PT now drains hella faster.

Vaibhav S.

From X · Aug 2026

You're going from 150 to 125. That's a 17% cut from what you have today.

Hossein

From X · Oct 2025

using every day and never get “approaching to weekly limit” and now just in a short time see this!

FlockTab

Subscription Agents track each login's 5-hour and weekly windows, and Limits over time shows the trend.

Your own subscriptions

Agents doing risky things

Sebastien G.

From X · Aug 2026

Claude decided to test a sandbox it was building by running `rm -rf` on my home directory

Bruno L.

From X · Jul 2026

GPT-5.6 Sol just deleted my whole production database.

FlockTab

Agent Guard holds deletes, production deploys, payments and email sends until you allow them.

How Agent Guard works

No record of who spent what

Ruth H.

From X · Sep 2026

If an AI agent runs up a bill on your account, the only record of what it did belongs to the company sending you the invoice.

Name withheld

From X · Sep 2026

Why did our AI bill raised by $7000 this month ?

FlockTab

One ledger across every Agent, key and provider: who spent what, where, and why a call was stopped.

See the whole flock

Quotes from public posts on X, Oct 2025 – Sep 2026. Names shortened; the authors are not affiliated with FlockTab.

How it works

Reserve. Allow or block. Close the tab.

Your agent

$ tab claude

FlockTab · my-projectCLOSED

$48.18of $50.00

✓ tab open
× cap
· velocity
· model
· tools

The provider

Anthropic · OpenAI · xAI

402 tab_closed · the provider is never called

Reserve before the call

Start the agent with tab, or point it at FlockTab's base URL. Each request reserves an estimated cost against the agent's tab.

$ tab claude
# or by hand
base_url = "https://proxy.flocktab.com/v1"
Authorization: Bearer ft_live_…

Allow or block, in the proxy

The tab, the cap, velocity, the model and irreversible tools are checked before the provider is called. Allowed calls pass through. Blocked calls never reach the provider.

Close the tab at the cap

At the hard cap the tab closes. The agent gets a 402 with a reason. Reopen it by hand, on a schedule, or never.

What you get

Control at the wire, not a report after it.

Hard caps per agent

Daily, weekly, or per run. Caps are enforced on reservation, so the last call that would cross the line is the one that gets blocked.

Kill switch

Close any tab, or the whole flock, from the console or one API call. Takes effect on the next reservation.

Close all tabs

Audit ledger

Every reservation, decision, and close is a line item with a reason, per Agent, project, model and login.

dec_8f21 · blocked · policy:cap · $0.44

Policies per Agent

Caps, velocity limits, model and irreversible-tool allowlists, and Agent Guard, set per Agent in the console or from the terminal. No policy file to write.

$ tab policy my-project --velocity 60
$ tab policy my-project --models claude-sonnet-5,gpt-5
$ tab policy my-project --guard ask
# irreversible tools stay denied unless you --allow them

Live decisions

Every call is a decision you can read.

The Ledger and Agents views stream what the proxy decided and why, and tab top shows it in your terminal. Filter by agent, policy, or outcome. Nothing is summarized away.

Activity · flocklive

14:02:35docs-site reserve$0.44BLOCKED

14:02:34my-project reserve$0.38ALLOWED

14:02:33billing-api reserve$0.12ALLOWED

14:02:31lead-enricher velocity 121/min$0.09FLAGGED

14:02:30eval-runner reserve$0.21ALLOWED

14:02:28docs-site cap reached$2,000.00CLOSED

Agent Guard

Every risky call waits for you.

Agent Guard checks your agents' tool calls before they run, in Claude Code, Codex and eight other harnesses.

Risky calls are held.: A force-push, a production deploy, a payment or a recursive delete is stopped by rule before it runs. Anything else it flags comes with the reason in plain words.
You decide, in one place.: Held calls wait in the console's review queue or in tab guard. Allow once, always allow, or keep blocked.
Safe work doesn't wait.: Reads, searches, plainly read-only commands and repeats of a call you already allowed go straight through. Set each Agent's level: Relaxed, Balanced or Strict. On Team and Fleet.

Start a flock How Agent Guard works

26 seconds, with sound. Captions on.

Agent Guard on 6 · loop watch on 2 · last 24h 184 checked, 5 flagged, 3 stopped, 0 stuck ────────
  14:02:11 api-server  DENIED  deny rm -rf ./backups                       Deletes files recursively · can't be undone · by rule
  13:58:40 web-app     ASKED   ask  git push --force origin main           Force-push · can't be undone · by rule
  13:51:07 docs-bot    ASKED   ask  aws s3 rm s3://docs-prod --recursive   Deletes S3 data · can't be undone · by rule

$ tab guard
jev_7f3a2c91-4b0e-4d2a-9c11-0e5d8b6a2f40  api-server-claude (claude) · pending
  rm -rf ./backups
  Deletes files recursively · can't be undone · by rule

Beyond the cap

One command in front of every agent, and everything it spends in one place.

Priced per agent. Not per token.

Full pricing

Solo

Free

3 named Agents · hosted or self-hosted

Hard caps and a kill switch
Velocity and model limits
14-day ledger
Passkeys and two-factor
Community support

Team

$19/ mo

10 included, then $2 per extra Agent

Launch offer: code FLOCKLAUNCH, 50% off your first month of Team monthly, until 9 Oct.

Everything in Solo
Irreversible-tool allowlists
Alerts to Slack, Discord, Teams
Outside spend and projects
90-day ledger and CSV export
Email support

Self-hosted

$0–$19/ mo

Solo or Team list. Keys stay on your machine

The proxy runs on your machine
Keys and prompts never leave it
The same caps, policies and ledger
Same price as Solo or Team

Fleet

$199/ mo

10 people, then $15 per person

Everything in Team
Unlimited Agents
Members, roles and audit log
SSO and SCIM
Unlimited ledger
Dedicated Slack support

Put a tab on every agent tonight.

Install tab. Set a cap. Sleep.

Start a flock Talk to us

Every agent runs a tab. Hit the cap, the tab closes.

13 coding agents run on a tab.

With Agent Guard

On a tab, without Agent Guard yet

A loop that retries forever spends forever. Alerts arrive after the money.

What people building with agents keep running into.

Surprise bills and overnight loops

No hard cap, no kill switch

Subscription limits that shrink

Agents doing risky things

No record of who spent what

Reserve. Allow or block. Close the tab.

Reserve before the call

Allow or block, in the proxy

Close the tab at the cap

Control at the wire, not a report after it.

Hard caps per agent

Kill switch

Audit ledger

Policies per Agent

Every call is a decision you can read.

Every risky call waits for you.

One command in front of every agent, and everything it spends in one place.

tab and tab top

Your own subscriptions

Agent Guard and loop watch

Outside spend

Projects

Observability

Priced per agent. Not per token.

Put a tab on every agent tonight.

Ask us anything.