Skip to content
New: answer your agents from your phoneLearn more

Make Claude Fable usage limits last longer.

Inspect the requests Claude Code sends to Fable, check cache costs before resuming a session and cut repeated explanations with Economy.

You return to a Fable session after lunch and ask for a small change. Claude Code sends the conversation along with your prompt, including the files it read and the commands it ran. If the prompt cache has expired, the provider charges to cache that history again.

We’ve added request inspection and a cache-expiry warning to ket’s main branch for the next public release, along with a reply style for shorter updates.

Claude Fable’s weekly usage limit

As of November 3, 2026, Anthropic includes Fable 5 and Fable 5.1 on Max and Ultra, and on eligible premium Team and seat-based Enterprise seats. You can spend up to 50% of your plan’s weekly allowance on Fable, within the overall allowance. On Pro and standard Team seats, you pay for Fable with usage credits from the start. Check Anthropic’s Fable plan rules for your seat type.

Run /usage in Claude Code to check your allowance. Use API dollar estimates to compare the cost of requests; Anthropic doesn’t publish a conversion from those prices into a fixed number of subscription prompts or hours. Our Claude Code usage limits guide covers the session and weekly windows.

Fable 5.1 cache misses and API costs

With prompt caching, the provider reuses an unchanged prefix of a request at a lower price. After the cache expires, you pay to write that prefix again. For a session with 300,000 tokens of history, you would pay these amounts at Anthropic’s standard Fable 5.1 API prices, checked November 3, 2026:

Input operationPrice per million tokensFor 300,000 tokens
Read from cache$0.25$0.075
Write a five-minute cache$12.50$3.75
Write a one-hour cache$20.00$6.00

You pay $6 to write that history to a one-hour cache, versus 7.5 cents to read it. These figures cover the existing prefix; you pay for new input and output in addition. See Fable 5.1 model pricing.

The provider keeps the default API cache for five minutes and offers a one-hour option. You refresh its lifetime by reusing the cached prefix. You can cause a miss by changing the prefix before it expires. Anthropic’s prompt caching documentation explains the rules.

Check the cost before resuming a session

Open a Claude session in ket and you’ll see a warning if its cache appears to have expired and it has at least 100,000 tokens to cache again. ket estimates the context size and cache lifetime from the last response in the transcript, then shows the cost at published API prices.

Continue if you need the history to finish the task. For unrelated work, choose Start fresh. You’ll get a new session beside the old one in the same worktree, and you can return to the old session with anything you typed intact.

You get the choice as you open the session on your Mac or phone. In Settings → Economy, you can change the token threshold or turn the warning off.

Use ket’s network inspector for Claude Code

Turn on Log what agents send in Settings → Network, then start a Claude Code session in a worktree. You’ll see requests from sessions you launch after enabling the setting.

  1. Open Network in the worktree’s right panel. You’ll see the newest requests first, with the amount sent and how much this turn added.
  2. Open a request in a tab to see what that turn added: your new message and the agent's tool results, plus the system prompt and tool definitions when they changed.
  3. Read the new content to find large file reads or repeated command output.
  4. Adjust the next run. Ask for a file or line range, trim command output, or start a fresh session for a separate task.

You might find that the agent added a long test log to a request. Trim that output before the next run. If the system prompt takes up too much space, shorten your instruction file. The inspector measures bytes; you need the provider’s token counts to check cache hits.

You can inspect Claude Code sessions that connect to Anthropic. ket leaves sessions using custom gateways, Bedrock, Vertex or Foundry on their existing routes, outside the inspector.

You opt into logging, and ket stores the request bodies on your Mac for five days. Copy or delete a request from its tab, clear a worktree’s log from the ⋯ menu, or delete logs in Settings. ket deletes the logs when you delete their worktree.

ket flags patterns that resemble secrets or personal identifiers and shows where the agent found them. You see these flags after the agent has sent the request. ket passes requests through unchanged.

Keep Fable’s replies short with Economy

You pay for Fable’s progress reports and summaries as output, then send them as history in later requests.

Choose Heavy or Max in the new standard Economy pack to ask Claude Code for terse engineering notes. You’ll get results first and shorter updates. ket includes the style instruction in the system prompt, where it persists through compaction.

You can use this style for chat while the agent follows your repository’s conventions for code and written files, including commit messages and pull requests. The agent must write warnings with enough detail for you to act on them.

Check the full policy before choosing a level. Heavy and Max change effort and subagent settings as well as compaction and command output. For a hard Fable task, you could spend more on retries than you save with a lower effort setting.

You can cut input with the command filter, which trims successful command output and preserves failures. See the Economy documentation for the settings each level applies.