CJ Avilla

Going to do my best to share helpful updates about how to get the most from the Claude Platform: APIs, SDKs, the ant CLI and more. If you have any feedback about existing API features you’d love to see, I’m all ears! Pumped to be back in a position where there’s so much I’m excited to use and hype.

🫰 tip: wanna pay 50% less for tokens in exchange for latency? Or maybe you're processing a ton of data? Checkout the batches API!

We've reset 5-hour and weekly rate limits for all users.

Slippery slope: claude “/schedule daily order 2 pints of Ben and Jerry’s Chunky Monkey“ … claude “plan a fitness routine to help me lose that DoorDash-CLI-15lbs” 😅

@dani_avila7 Bingo. :)

@cjav_dev I was just talking with a coworker about what happens to the cache if you change the CLAUDE.md file in the middle of a session... Well, here's the answer

🫰tip: you don't need to *manually* manage your prompt cache breakpoints. Put this one liner at the top level of your /v1/messages call to automatically cache prompts save some cash!!

@dani_avila7 Claude Code handles prompt caching for you unless you disable it. 👍👍 code.claude.com/docs/en/prompt…

@cjav_dev That’s great 👏 Is there a way to configure prompt caching behavior in Claude Code, or does the CLI handle cache breakpoints automatically with no user control?

Why is brainstorming and riffing on the future so damn energizing?

Notebooks to learn the ant CLI for Claude Managed Agents I spent the weekend running Claude Managed Agents from the terminal, so I turned it into an open-source cookbook Jupyter notebooks you run cell by cell and actually READING THE CODE! Every topic ships twice, first with the Python SDK, then with the ant CLI, so you see the same flow from both sides. Two lessons so far: - Basics: environment, agent, session, and the stream-first pattern - Multi-turn: one session reused across turns, history stays server-side, you never resend prior messages

@tomcoustols @ethanolivertroy @Laz4rz @grok 1Password and dev environments

@ethanolivertroy @Laz4rz @grok Any recs?

So apparently @grok is just sending your .env secrets straight back home

🫰 tip: Task budgets are great if you have predictable per-task cost. Let Claude self-regulate spend within a budget on long-horizon tasks.

one of the most impactful memory usage optimizations in bun: schedule gc when the http server finishes sending a response body not clear what the equivalent is for CLIs

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

@VladAndries13 @ClaudeDevs Sorry about that. We're working on a fix!

@ClaudeDevs I tried this setup, but every time Sonnet tries to call Fable, it receives Advisor unavailable. Any fix for this?

A few patterns we frequently use with Fable 5: Use Fable 5 as an "advisor." An executor (Sonnet 5) calls Fable 5 for guidance. Most tokens are billed at the lower executor rate.

@JavierRomeroCR3 @ClaudeDevs sorry this is a known issue and we're working to fix!

@ClaudeDevs /advisor no works with opus and fable

Claude Code on desktop now has an in-app browser. Claude can pull up docs, designs, or any other site. It can read, click through, and interact the same way it does with your local dev servers. It's sandboxed and configurable: you choose whether sessions persist.

@richkuo7 @ClaudeDevs In cell 3 in this cookbook, you'll see there's a worker agent being created, then it's ID is passed in when creating the coordinator agent: github.com/anthropics/cla…

@ClaudeDevs would love to see a practical example, and if this is something that can be prompted into claude code

Claude Managed Agents supports both patterns with sub-agents. Escalate up to a Fable 5 advisor, or delegate down to Sonnet 5 workers. Each sub-agent keeps its own cache, so repeat calls don't pay in full for the same context twice. platform.claude.com/docs/en/manage…

@richkuo7 @ClaudeDevs When you create your agent, you can pass in a param "multiagent" with a roster agents that should work together with this agent. #create.multiagent" target="_blank" rel="nofollow noopener">platform.claude.com/docs/en/api/cl…

@ClaudeDevs not sure i understand this correctly, i have to build my own multi agent session? or is this something i can prompt? the docs don't really give good examples or use cases

#key-expiration" target="_blank" rel="nofollow noopener">platform.claude.com/docs/en/manage…

You can now set an expiration when you create an API key or an Admin API key in the Claude Console. That said... I've been using the ant CLI for auth instead of API keys. Login with `ant auth login` and then the SDKs can load the credentials implicitly.

I have a few Claude managed agents doing daily chores in a sort of "headless" mode. - No UI (I can look in platform.claude.com/agents if i need to see a session). - No glue code or wiring, no webhooks or anything to kick them off Each one is a scheduled deployment: it wakes up on cron, starts a session, hits a few APIs and MCP servers, does its work, and builds up memory across runs. I expect to see more and more of these headless Claude Managed Agents quietly chipping away at tasks.