Adapt

Adapt offers company agents that teams can hand work to, working in tools such as Slack, email, and the web, and acting with the access of whoever asks. It also provides agent infrastructure that can be used together or separately with your own agents: an open-source MIT-licensed harness for running the agent loop in your repository, sandboxes for running code per job with metered billing, storage for files and memories shared across agents, and a model gateway offering one API to major models with your own provider keys at 0% markup or Adapt's keys at list price plus 5.5%. The page lists a harness developer preview dated January 4, 2027 and a waitlist, and includes customer quotes from Wander, DoNotPay, and Orion Sleep.

Key features

• Company agents that answer in Slack, email, and the web and act with the access of the requesting user
• Personal and team agents, including examples for support triage, renewals, month-end close, security reviews, and onboarding
• Open-source MIT-licensed agent harness that runs in your repository, with instructions and tools kept as code
• Ability to swap in your own models, sandboxes, and storage
• Sandboxes with up to 32 vCPU and 32 GiB memory, metered per second while running, and parked sandboxes that are not billed
• Storage for files and memories that persist after a sandbox stops and can be shared across agents
• Model gateway with one API, configuration-based model switching, and the option to use your own provider keys at 0% markup
• Support for using a different model for each step of a workflow

Best suited for

• Companies that want AI agents integrated with their existing tools and knowledge
• Developers who want to build custom agents using individual infrastructure components
• Teams wanting to run code-based agent tasks in isolated sandboxes
• Organizations that want to use multiple AI models through a single gateway

Pricing

Not stated on the website