how it actually works:
every agent gets a score — output completeness × eval pass rate × frequency × trend. Pause/block agents that drop below your threshold, and set the limits you need. No runaway spend, no manual cleanup.
after each session, Monitor shows you the timeline — which steps actually ate your context and where Akai spent too long thinking.
Then it fixes the workflow for you: clarifies the prompt, scopes the tools (that 98kb fetch becomes 6kb), caches to optimise and collapses repeat steps into a script — even making suggestions based on best practices that it's seen elsewhere.
the loop gets cheaper and sharper every run. Weak agents get cut, the strong ones compound.
most teams have no idea where their agent token spend actually goes — or how to control it.
we just shipped Akai Monitor to fix that. After every run it shows you exactly what's burning budget — then fixes it. Tighter prompts, scoped tools, repeat steps turned into scripts
granular token visibility, and control — finally.
Claude Fable 5 is now available in Akai.
Choose the right model for every agent. Akai lets you pick from multiple AI providers and models, so you can optimize each agent for its specific task - balancing speed, quality and cost.
Give it a try 🔗 https://t.co/dZ0kcLywuJ
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
Its capabilities exceed those of any model we’ve ever made generally available.
Hey Patrick, we're building this with Akai. Which now runs agent workflows end-to-end across client organisations (including Deel's).
This is all live in Akai:
- Input files & context → store static docs, connect internal wikis, Confluence etc as persistent agent context that it refers to at runtime
- Real-time collab → we have real-time collaboration where your team can collaborate on the same workflows at the same time, with RBAC, VCS, rollbacks and audit trails
- Stored prompts & workflows → handled at the Agent level (workflows can be re-used, shared, duplicated) and at the prompt level via Skills (to re-use part of your workflow for other agents)
- Beyond chat →Akai is multi-model and executes code, calls endpoints, discovers unofficial APIs, automates browser interactions, and orchestrates multi-step workflows.
- Shareable outputs → agents, workflows, and outputs are all storable, shareable, and API-callable
Would love to show you
At Akai, we realised that enterprise-grade agents should be built by the people who actually know the processes.
So we made it easy to do just that.
Show Akai your workflow once, and Akai learns every step, builds the connectors and creates the agents automatically with all the context it needs. Your team stays in control throughout.
Get early access → https://t.co/0gPd5dVVLN
All of Deel's operations, running on Akai🕶️
Come along to our webinar tomorrow, to find out how you can join the fun
https://t.co/9WHrLITHRS
https://t.co/dZ0kcLz4kh
100% of Deel's ops teams run on @Akai_by_deel.
We didn't beta test Akai on customers, we ran it on ourselves first.
Now it’s your team’s turn.
See it in action at our webinar May 28 → https://t.co/x9FvDV3uNH
I love it when I hop on an @akai_by_deel demo.
Everyone starts skeptical. "We are building this in house."
100% of the time, 15 minutes later, they're beliebers.
@LexSokolin@Bouazizalex@LexSokolin typically the people in large organisations who are doing manual, repetitive work across internal & external systems. Finance, ops, legal etc. - we build directly with the SMEs.
Akai learns all about your business and builds interconnected agents to do the work.