Ship the agents you already built

Nine in ten agent pilots never reach production — not because the model was wrong, but because nobody could say what the agent was allowed to do, what it actually did, or what it would cost at scale. A Kohala agent is built from six parts: agent scripts that do the work, managed memory, a schedule, the live Koans it publishes, alerts, and governance — a tool allowlist the runtime enforces before a call reaches a tool, token caps of 25,000 per run and 500,000 per day by default, and a run record of the tool calls the runtime brokered, the validator verdicts, tokens and seconds it spent, plus the charter, validators and caps you set — so every run leaves a record you can hand to whoever blocked it.

Operational control for enterprise agent fleets

Kohala helps enterprises improve Efficiency by identifying failed and repaired runs and wasted execution; lower Risk by exposing spend variance, failures, missing limits, fallbacks and governance gaps; and increase Determinism by measuring repeatable validated execution and unnecessary live-model dependence.

  1. Build + Deploy on Kohala through Kai, with the hosted runtime applying configured controls.
  2. Build Anywhere + Deploy on Kohala by bringing Python or TypeScript work into hosted governance through Kai and import surfaces.
  3. Build Anywhere + Deploy Anywhere through SDK and DevKit connected reporting. Kohala governs the evidence it receives; off-platform side effects remain locally enforced.

Three ways in. All three end with a governed agent.

  • Bring yours — point Kohala at a Python loop (LangChain, LangGraph, CrewAI, the OpenAI Agents SDK, or PydanticAI) or a TypeScript one (LangGraph.js). Nothing is rewritten; your loop becomes the agent's scripts under the same charter and tool allowlist as everything else, the run record appears on the first run, and you add the schedule, memory, koans and alerts when you want them. n8n stay where they are and report each run instead. Start in the import wizard or import with Kai.
  • Describe one to Kai — say what to watch or do, and Kai scores your data sources, writes the agent scripts, and sets up whatever the job needs — a schedule, koans to publish, alerts — then drafts the charter and spend cap and ships all of it on a single Approve & Build. Free, no account.
  • Remix from the catalog — working agents built by the Kohala team, by industry and by tier. Each arrives already configured, with its scripts, its schedule, and the koans and alerts its job needs, and Remix opens it in a conversation with Kai you can change. No account needed. Free and premium.

Every door produces the same kind of agent — six parts

A remix or a Kai build arrives already configured. An imported loop starts as the agent's scripts, under governance, and picks up the rest as you add them.

  • Agent Scripts — the work is code the agent runs, Python or TypeScript, written by you or by Kai. Not a prompt improvising a different answer each time.
  • Memory — a managed store the agent can read and write across runs: prior outputs, source samples, the baseline it compares against. No database to stand up.
  • Schedule — a cadence, not a chat window. A scheduled agent wakes on its own, does the work, and has finished before anyone thinks to ask.
  • Koans — the live cards it publishes: status displays, trend charts, input forms. Its runs can refresh them, and they embed where your team already looks.
  • Alerts — rule-based notifications to named recipients, by email or in-app, so you hear about the exception rather than reading a log nobody opens.
  • Governance — a tool allowlist the runtime enforces before a call reaches a tool, token caps of 25,000 per run and 500,000 per day by default, and a run record of the tool calls the runtime brokered, the validator verdicts, tokens and seconds it spent, plus the charter, validators and caps you set.

Start with one agent. Grow into a governed fleet.

  • Free — run records and approval gates on every agent, charter and allowlist editing, free reference implementations, scheduled runs and snapshot links.
  • Professional — team seats, approval routing, integrations, embeddable Living Koans, and the four business agent packages: Sales & Customer Relations, Growth & Marketing, Finance & Accounting, and Inventory & Operations.
  • Enterprise — roles and access control, SSO, SCIM, exportable audit logs, your own S3, Pinecone, and model keys, and MCP fleet control.

Built for three kinds of people

  • Builders — any framework in, governed out, paid on every run.
  • Operators — put a supervised workforce of agents on your operation.
  • Partners — ship a governed AI offering without building one.
  • Platform — the governance, memory, and monetization fabric for AI agents.