caliber is the intelligence layer for how your org runs on AI. plug in your HRIS and see: - AI spend by team and by owner - $ per shipped item, per team - every seat, key and agent, plus unused ones already live with paying customers. trycaliber.ai

lot asked so

Happy to partner with @nvidia, details soon :)

Start letting Cal unpack your AI bill, down to production value and individual ownership.

putting a cap on tokens is easy. knowing which team burned through it, what shipped, and what broke after launch is the hard part. a cheaper model won't fix a workflow nobody owns.

careful..

ai adoption is a terrible north star. a team can double its prompts, tokens and active seats without shipping one extra thing. show me accepted work per team, what it cost, and who owns the bill.

connect HiBob, Jira and Anthropic. get this view in 3 minutes, with queryable insights and actions. AI spend and tokens for every team and manager, rolled up the org chart, plus a year of spend, PRs and cycle time next to it. this is caliber v2. @HiBob_HR @Jira @AnthropicAI

the free part of caliber. ai-setup writes CLAUDE.md, cursor rules and AGENTS.md from your repo, keeps them current on every commit, syncs skills across agents and compacts long sessions with Jev. 1,280+ stars, 2.4K npm installs last month. github.com/caliber-ai-org/ai…

your HRIS already knows the org. caliber adds the AI layer. connect HiBob, tag teams with their tools, and get spend, usage and output per team, with the actions to fix what's off.

divide the ai bill by people who actually used ai, not by headcount. a team of 4 and a team of 40 become comparable, and the outliers stop hiding inside the org total. that one change answers more budget questions than any seat report.

claude code now keeps running with your laptop closed. great for throughput. it also means a growing share of your ai bill is work nobody watched happen. who started each session, what it was for and what it shipped is about to matter more than seat counts.

ask any eng org how many api keys they have and you'll get a guess. the real list is usually worse: most enabled keys made zero calls last month, a chunk never expire, and nobody remembers who owns half of them. key count isn't posture. owner, last use and expiry are.

opus 5.5 and gpt-6 both got cheaper this week. your ai bill probably won't. cheaper tokens mean more agents, longer runs and more retries. list price is the menu, not the bill. the number worth watching is cost per finished task, by team. that's the one that tells you if the price cut actually reached you.

most ai budgets are still built around vendors. that's backwards. the unit that matters is work: what job ran, who owned it, which agent handled it, and whether the output shipped. tools change every month. the accountability layer shouldn't.

seat count is a terrible benchmark. two teams of ten can look identical on paper while one has 90% active usage and the other has two people carrying the bill. compare cost per active person, then ask what changed in the work.

the first step in controlling ai spend isn't cutting tools. it's making the bill legible by team. once ownership is visible, adoption, idle seats and outcome questions stop living in separate spreadsheets.