Oncell

Oncell is an enterprise agent platform for building, testing, evaluating, running and improving AI agents on a single runtime. Every agent action passes through one boundary that serves as the trace, the test, the reward and the audit trail. An agent is defined in one TypeScript file containing its instructions, model, tools and a spending budget. Memory, a database, files, human approval and schedules are built into the runtime. For testing, a customer's agent can be forked into a staging copy in seconds and tested against real state without touching the original. Evals are declared beside the agent using rubrics or assertions, a scorecard is produced for each run and compared with the previous one, and regressions block the release. The run log records every action, approval and cost per customer, and budgets are enforced below the model level. Run logs and evals are also used to search prompts and models, with post training of customer owned models described as the next step. It is aimed at engineering teams running production agents for many customers, such as a platform that gives each small business its own agent. A published example meters usage per business at $0.0006 per customer conversation, and the site says an agent costs nothing while it waits on a human or a schedule. A full pricing page is not published. Interested teams are directed to talk to the company.