Guide from the official docs
Hit a usage limit on your AI agent? What each one counts and when it resets
What 11 AI agents count toward usage limits, when the allowance resets, what happens when you run out, and where to check what's left, per vendor docs.
This guide follows the official docs and announcements, linked inline. AideDex did not test these products and does not score them. How we check facts.
Last checked Oct 24, 2026; LVLZ and Claude plans re-checked Oct 31, 2026.
What a usage limit means depends on how your agent counts work. Some agents share your chat plan’s limits, like Claude Cowork and Gemini tasks. Some have their own allowance, like Grok Bot, Poke, LVLZ and Meta Muse. Credit agents like Manus, Perplexity Computer and Lindy spend from a credit balance. Bring-your-own-key agents like rabbit OS3 have no vendor limit at all: your model provider’s account is what runs out. Find your agent’s group below to see the unit, the reset window, what happens at the limit, and where to check what’s left.
To estimate monthly spend across plans, use the cost calculator.
Quick reference
| Agent | Counting type | Reset | Where to check |
|---|---|---|---|
| Claude Cowork | Shares Claude plan limits | 5-hour session + weekly | Settings > Usage |
| Gemini tasks | Shares Gemini Apps limits | Every 5 hours, up to a weekly limit | Not stated on the limits page |
| ChatGPT Work | Shares Codex usage | See Codex pricing | Not stated on the Work article |
| Grok Bot | Own allowance | Weekly | Not stated |
| Poke Free/Pro | Own allowance | Refills automatically | Subscription page |
| Poke Ultra | Credits ($1 each) | Monthly billing cycle | Subscription page |
| LVLZ | Own allowance (no message limit) | Monthly: the 1st on Free, at renewal on Autopilot | Web app (Autopilot only) |
| Meta Muse | Own allowance (Muse tokens) | Weekly | Not stated |
| Manus | Credits | Monthly, on subscription date | Not stated in the credits article |
| Perplexity Computer | Credits | Monthly | perplexity.ai/account/usage |
| Lindy | Credits | Monthly | Not stated on pricing |
| rabbit OS3 | Your model provider | Your provider’s terms | Not stated by rabbit |
Agents that share your chat plan’s limits
With these agents, agent work comes out of the same limits as normal chat. For Claude, our Claude Cowork setup guide shows where Cowork fits in each plan, and the Claude Cowork page lists the plans that include it.
Claude Cowork
Anthropic says multi-step tasks “like ones that run code, create files, or use your connected apps and browser, use more of your usage than a quick question” (Anthropic Help Center). Tool use, model choice and artifacts also affect how fast you use up your limits (Anthropic Help Center).
- Reset window: On Pro, Max, Ultra, Team and seat-based Enterprise plans, there is a five-hour session limit and a weekly limit; verified Ultra accounts have no five-hour limit. Under Current session, Anthropic shows how much of the five-hour limit you’ve used and the time left in the session. Weekly limits shows when the weekly limit resets (Anthropic Help Center).
- Where to check: Settings > Usage, under Plan usage limits (Anthropic Help Center).
- More usage: Paid plans can turn on usage credits. Usage-based Enterprise plans have no set limits and are charged by consumption (Anthropic Help Center).
- Saving usage: Project content gets cached and “counts less against your limits when reused” (Anthropic Help Center).
Gemini tasks
Google says “Gemini Apps have compute-based usage limits” that factor in “the complexity of your prompt, the models and features you use, and the length of your chat”. The same page lists Tasks in Gemini among the features covered, for AI Pro and AI Ultra (Gemini Apps Help).
- Reset window: “Your limit refreshes every 5 hours until you reach your weekly limit” (Gemini Apps Help).
- Upgrade path: AI Plus gets 2x the standard limits, AI Pro 4x, and AI Ultra 5x or 20x the AI Pro limits, depending on the subscription (Gemini Apps Help).
- What uses more: Some features “(like media generation or Deep Research) will consume more of your usage”. Google suggests starting a new chat so Gemini doesn’t re-read a long history (Gemini Apps Help). Setting up the agent itself is in how to get started with Gemini tasks.
ChatGPT Work
OpenAI says “Work follows the same usage structure as Codex”. It points to the Codex pricing page for included usage and extra credits (OpenAI Help Center).
- Shared pool: Tasks started through Voice “draw from the same shared agentic usage and credit pool as Work and Codex” (OpenAI Help Center).
- Long tasks: “Longer conversations and tasks still count toward your usage limits” (OpenAI Help Center).
- After the allowance: Plus and Business Standard seats include limited Astra usage, “with optional credits for additional usage afterward”. On Pro $500, Ultrafast uses included usage at 8× the Standard rate, then draws credits at a rate 6× that of Standard credits (OpenAI Help Center).
- The Work article doesn’t give a reset window. It sends you to the Codex pricing page for that.
Agents with their own allowance
These agents have their own allowance, separate from any chat plan. The Grok Bot page and Meta Muse page list the plans and countries each one covers.
Grok Bot
xAI says “Grok Bot subscriptions come with weekly usage included, with additional usage billed based on token cost”. Higher tiers list “3.5x Premium usage” and “12x Premium usage” (xAI). xAI doesn’t say on that page where you can see remaining usage or what message you get at the limit. Plans and first steps: how to get started with Grok Bot.
Poke
Poke says every plan has an allowance, and “Poke keeps working when you reach it” (Poke Docs).
- Free and Pro: When you hit the limit, Poke keeps working in “a lighter mode until it resets”. On Pro, “the most demanding work waits until your allowance resets”. The allowance “refills automatically” (Poke Docs).
- Ultra: Ultra uses “a monthly pool of credits (1 credit = $1 of usage)”. The pool refills “at the start of each monthly billing cycle”. If you run out, Poke “automatically continues at the full Pro tier experience”, or you can buy more credits from the subscription page and turn on auto-reload (Poke Docs).
- Where to check: “Your current usage and reset date are always shown on your subscription page.” Poke says heavy usage often comes from automations that run very often (Poke Docs).
Setup steps are in how to get started with Poke, and which plan unlocks email in Poke and the other agents is in AI agents for email by plan.
LVLZ
LVLZ’s docs say “No message limits” as of Oct 31, 2026. Instead, one fair monthly allowance covers messages, voice notes, calls, browser tasks, files and connected accounts (LVLZ Docs).
- Calls and other heavy work: calls are capped at 5 minutes a month on Free (after the workspace sees real use) and 30 on Autopilot, with each call lasting up to 15 minutes. Browser tasks draw from the allowance for each minute the browser runs, and each connected account on Autopilot draws a little every day (LVLZ Docs).
- At the limit: the agent switches to a lighter mode and keeps replying. It can still read and explain your clients, bookings and numbers, but changes nothing until the month renews, on the 1st on Free and at your renewal date on Autopilot. Routines, workflows and check-ins pause; reminders still come through, and the web app and CRM work as normal (LVLZ Docs).
- Where to check: on Autopilot the web app shows usage as a percentage of the plan and warns you near the end. Free shows no meter (LVLZ Docs).
- More usage: the pricing page sells no usage top-up. The only add-on is a connected account beyond the 3 in Autopilot, at $10 a month each (LVLZ pricing).
Meta Muse
Meta measures Muse in tokens per week. The Power plan includes 500M Muse tokens a week and Maximum includes 3B. “Weekly usage allowance for Muse increases as you upgrade plans.” On the free tier, once you reach the limit you “can upgrade to a paid subscription or wait until your free usage limit refreshes” (Meta Help Center). For availability outside the US, see Meta Muse outside the US.
Where to pick a plan and how sign-in works step by step: how to get started with Meta Muse.
Agents that run on credits
Credit agents spend from a balance. How much a task costs depends on what the agent does. Plan prices for Manus and Perplexity Computer are on their pages.
Manus
Manus credits are “primarily consumed” by LLM tokens, virtual machines and third-party APIs. They’re “only consumed during active task processing” (Manus Help Center).
- Reset: Monthly credits “automatically refresh on the same date each month based on your subscription date”. Unused monthly credits reset and don’t roll over. Add-on credits carry over as long as your paid subscription remains active (Manus Help Center).
- Order of use: event credits > daily credits > monthly credits > add-on credits > free credits (Manus Help Center).
- At zero: Manus bills every action when the following action begins, so “your credit balance may temporarily show a negative value”. Credits are refunded for tasks that fail because of a technical problem on Manus’s side (Manus Help Center).
- More credits: Upgrade to Pro, or switch to a higher plan (Manus Help Center).
Perplexity Computer
Max subscribers get 10,000 credits a month. Consumer Pro plans include no monthly credits. Credits are “used only for Computer tasks”, and Ask searches stay unlimited (Perplexity Help Center).
- At zero: “If your credits run out, active tasks pause (not cancel) and resume automatically when credits are available.”
- Top-ups: Auto-refill can purchase extra credits once your balance drops under 500 credits (Pro) or under 2,500 (Max). Monthly spending limits are also available.
- Where to check: perplexity.ai/account/usage (Perplexity Help Center).
Running Computer on your own Mac or PC has its own plan and hardware rules; see can you run Perplexity Computer locally?.
Lindy
“Credits measure the work Lindy does. You only use them when Lindy is working.” Lindy gives these ranges: “Everyday Asks” take 2 to 250 credits and “Deep work” takes 250 to 1,000. On Oct 30 the pricing page showed some visitors plans from 3,000 credits a month and others from 5,000, and credits are pooled across your team. If you run out, Lindy “pauses credit-using actions until your credits reset at the start of your next cycle”. Admins can “top up at $10 per 1,000 credits” (Lindy pricing). What a Lindy phone call costs in credits is in how to get started with Lindy.
Agents that run on your own model provider
rabbit OS3
“OS3 runs on an API key you bring from a supported model provider”, and “you pay that provider directly for what you use”. OS3 itself is free. OS3 checks each key when you save it. It refuses a key the provider rejects, “including one on an account with no credit left”, and tells you why. You can switch keys in settings → api keys (rabbit support). The article doesn’t say where to check usage, so rate limits and balances follow your provider’s own terms. Adding the key is a step in how to get started with rabbit OS3. For self-hosted agents with the same setup, see OpenClaw and Hermes Agent hosting cost.
If the agent stops for another reason
If your usage page shows allowance left but the agent still won’t run, the problem may be an outage, not a limit. See Is my AI agent down?.
Sources
13 pages- Anthropic Help Center: Get started with Claude Cowork support.claude.com
- Anthropic Help Center: Usage limit best practices support.claude.com
- Gemini Apps Help: Gemini app access and limits support.google.com
- OpenAI Help Center: ChatGPT Work and Codex help.openai.com
- xAI: Grok Bot x.ai
- Poke Docs: Usage & Resets poke.com
- LVLZ pricing lvlz.ai
- LVLZ Docs: Monthly usage docs.lvlz.ai
- Meta Help Center: Muse subscription plans meta.com
- Manus Help Center: Rules for credits consumption help.manus.im
- Perplexity Help Center: What is Computer? perplexity.ai
- Lindy pricing lindy.ai
- rabbit support: How to bring your own key (BYOK) in OS3 rabbit.tech