Changelog
What changed in FlockTab
Every change you can see, newest first: the tab CLI, the console, the proxy, Agent Guard and billing.
Follow it in a feed reader: /changelog.xml. Update tab with tab update; it shows the same notes.
tab CLI 0.1.43
- Agent Guard on Grok, Kimi Code, Pi and Cline too.
- tab update no longer restarts a proxy into fewer keys.
tab CLI 0.1.42
- Agent Guard: a refused call stays refused, even when held for review.
- tab guard allow <hold id> lets one held call through.
tab CLI 0.1.41
- tab pool login <name> signs a pool login in again.
- The pool skips logins that may have expired; forks stay forks.
- tab policy --all --guard watch, and --default-guard for new Agents.
tab CLI 0.1.40
- tab grok and tab kimi run on the flock's xAI and Moonshot keys too.
- Agent Guard sensitivity: tab policy --guard-level relaxed|strict.
- Agent Guard answers read-only and repeated calls at once.
tab CLI 0.1.39
- tab grok and tab kimi run on your own xAI and Moonshot keys too.
- tab cline is back, metered; Gemini CLI and agy tools are read.
- Agent Guard starts no hook for read-only tools: faster reads.
tab CLI 0.1.38
- Agent Guard (the new name) says in plain words why it stopped a call.
- Recursive deletes stop by rule; codex exec is guarded after Trust all.
- grok and kimi need a Subscription Agent; tab cline waits to be metered.
tab CLI 0.1.37
- tab gemini and tab agy (Antigravity) run on your own Gemini key.
- tab openclaw and tab cline; tab codex uses your own OpenAI key.
tab CLI 0.1.36
- Caps hold to the cent: input is weighed the way tokenizers count it.
- A busy proxy is no longer mistaken for a dead one.
tab CLI 0.1.35
- A call you hang up on settles at what it really used.
- Today's models are priced.
tab CLI 0.1.34
- A call is fitted to what is left on the tab, not refused outright.
- OpenRouter model ids are priced; the tab closes only at the cap.
Agent Guard
- Agent Guard checks tool calls on ten harnesses: Claude Code, Codex, Gemini CLI, Antigravity, OpenCode, Goose, Grok Build, Kimi Code, Pi and Cline.
- Pick how cautious it is per Agent: Relaxed, Balanced or Strict.
- A review page in the console: held calls as cards grouped by day. Allow once, always allow that exact command, or keep it blocked.
- It learns from your answers in Claude Code. Say no and that exact command is flagged at once; say yes and nothing changes until you trust it in the console.
- Edit every Agent's guard in place on the Agent Guard page, or set a value for all of them at once.
- Read-only and repeated calls are answered at once, and verdicts are in plain words.
- Agent Guard is on the Team and Fleet plans, with a monthly allowance of judged calls.
- Fixed: a refused call held for review could run when the hook stopped waiting.
Billing
- All 13 supported harnesses bill on a tab, on your own key or the flock's key, and each was checked against the vendor's own token count.
- Grok and Kimi Code run on your own xAI or Moonshot key, or on the flock's.
- Caps hold to the cent: a call is fitted to what is left, and a call you hang up on settles at what it really used.
- OpenAI's GPT-6 models are priced.
Console
- A short onboarding video when you first sign up.
- A next-step strip under the header until your first real Agent runs.
- Alerts: Describe it asks when a request could mean two things, instead of guessing.
- Outside spend by project as a ranked list, with lines that add up to the total.
- Error reports reach us through Sentry, scrubbed before they leave. On your own machine they are off unless you turn them on.
Site and docs
- The home page's Agent Guard section has a 32-second video: a production deploy held, allowed once in the console, then live. It plays when you press play, with captions on.
- Supported harnesses: one table in the docs and on the home page, showing each billing lane and Agent Guard.
- This changelog, with an Atom feed.
tab CLI 0.1.33
- API Agents keep the harness's own keys: tab only moves where calls go.
- tab opencode, pi, goose, aider and copilot run on the tab.
- A failed or empty stream is refunded, not charged.
Billing
- API Agents keep the keys the harness already uses; tab only moves where the calls go.
Proxy
- A failed or empty stream is refunded instead of charged at the estimate.
Site and docs
- A new home page, and every page runs text at full width.
- Docs: how to test an API-key Agent and switch a folder's Agent.
tab CLI 0.1.32
- tab top: a login at the pool's guard reads "at its limit".
tab CLI 0.1.31
- tab top: a banner at the top when a newer tab is out.
tab CLI 0.1.30
- tab update: Tally dresses for a dark terminal, like in tab top.
tab CLI 0.1.29
- tab <command> help, with examples for every command.
- --create and other switches no longer swallow the next word.
tab CLI 0.1.28
- tab top: #PR in front of each branch, read with gh on your machine.
tab CLI 0.1.27
- tab codex starts on Codex 0.156 with a self-hosted proxy.
tab CLI 0.1.26
- tab top: spend per branch, with the share of each login's week.
- ×2 when two terminals run one Agent; t dresses Tally for dark.
- Subscriptions line up across Claude, ChatGPT and Grok.
tab CLI 0.1.25
- tab top: sessions per Agent, and a readable Subscriptions panel.
- Tally wears a lighter coat on a dark terminal.
Console
- Spend per branch: each call lands on the branch its session was on, with the branch's pull request.
- A FlockTab MCP server: read spend, close a tab or change a cap from Claude Code, Codex, Cursor or claude.ai. Sign in with OAuth.
- CSV exports.
- Policies as one table: a row per Agent, a column per enforced setting.
- The loop watch can pause an Agent after it is stuck for a number of checks in a row.
- Outside spend: AWS, Google Cloud, Azure, Supabase, Render and Semaphore connectors, and bills shared by what the Agents did.
- Alerts for this month and this year, and Describe it to fill the form from a sentence.
- Fixed: security issues found in an audit.
Site and docs
- Pages for finance, managers, engineers, leadership and security, and a page for each feature.
- Public MCP setup docs, and llms.txt.
- Terms of Service and Privacy Policy; a contact form.
tab CLI 0.1.24
- The guard stops ssh, sudo and a sandbox turned off; tab policy --flag.
- Caps reset each window; tab cap --on-cap pause, --reopen window.
- tab cap <agent> --reopen works without an amount.
tab CLI 0.1.23
- tab top scrolls, and has a leaderboard.
- ChatGPT banked resets and Grok weekly credits.
- Smarter Agent names, and an animated tab update.
tab CLI 0.1.22
- tab top: the flock live in your terminal.
- Fable's 7-day window in the pool, tab accounts and the status line.
- Proxy 0.1.17.
Billing
- Fleet is priced by the person: $199 a month for 10 people, then $15 each.
Console
- Enterprise: members, roles and permissions, SSO, SCIM, project-scoped members and passkey sign-in.
- Passkeys and two-factor sign-in on every plan.
- Alerts on a tab, an Agent, a project or a slice of outside spend, to email, Slack, Discord or Teams.
- Cap windows: a cap per day, week or month, what happens at the cap, and when a closed tab reopens.
- The guard and the loop watch, set per Agent from tab or the console.
- Open all tabs, the other half of the Close all kill switch.
- An Overview of the flock right now, with tabs as cards or a table.
- Support from the site and the console.
Proxy
- Fixed: a retried subscription call could loop on ledger_unavailable.
Site and docs
- The /tab page, and 25 guides on the blog.
Billing
- Named Agents stop at the plan's limit, and Team adds Agents on the Plan page.
Console
- Model and reasoning effort on every call, on every ledger row, and a Models page.
- One harness per Agent, shown on every table.
- The observability map, and outside spend with date ranges and filters.
- Outside spend syncs itself every hour, and shared costs can be split across projects by weight.
- Agents link themselves to the project they are named after.
Proxy
- Codex, and harnesses with hundreds of tools, go through the tab.
- Grok Build on SuperGrok goes through the tab.
- Claude's Fable week is read as its own quota window.
Billing
- Subscription Agents: Claude Max, ChatGPT, SuperGrok and Kimi logins pass through the tab, recorded at list price and never charged to the cap.
- Cached input is priced.
Console
- Live tab spend without a refresh, and a subscription's use of each plan.
- The console Docs: start here, how it works, the tab CLI, subscriptions and the pool, self-hosted, troubleshooting and a glossary.
- Edit a tab's cap from its page.
- An Expo (EAS) connector for outside spend.
Proxy
- The self-hosted proxy: one Rust binary from npm. Keys stay on your machine; the ledger stays on flocktab.com.
Billing
- Provider keys are sealed with one unlock per flock, and it can be changed without pasting a key again.
Console
- The console: sign in, add provider keys, name Agents, set caps, and watch every call.
- Outside spend: GitHub, OpenRouter, Railway, Vercel, Cloudflare, ElevenLabs and OpenAI connectors, attributed to projects.
Proxy
- The proxy serves OpenAI Chat Completions, OpenAI Responses and Anthropic Messages, and streams replies through.
- A provider error refunds the hold; only a successful call is charged.
Proxy
- FlockTab's first version: a hard cap per agent, enforced before the provider is called.