Buster

Buster, now operating as Keystroke, offers an all-in-one platform for building internal AI agents and AI systems. Agents and workflows are authored as ordinary TypeScript in the customer's own repository — inputs defined with Zod schemas, tests run with Vitest — and then deployed to a hosted web application that the rest of a team can use. Deployed agents and workflows automatically surface in that app with a chat interface, workflow visualizations, input forms, sharing, credential management, integrations and run logs, so teams do not have to assemble streaming, permissions or run history themselves.

The platform's agent runtime wraps Vercel's AI SDK and gives agents persistent memory, a persistent file system with code execution in an isolated VM, a persistent signed-in browser, web search, database access, scheduled and event-based triggers, and optional VM sandboxes. Agents can call workflows, actions, other agents, MCP servers and integrations as tools; the company advertises access to 1,000-plus integrations, any API and any MCP server, and support for 300-plus models including bring-your-own API keys. A managed runtime is described as making workflows durable across errors, sleeps, human approval steps and process restarts. Agents can be used through the Keystroke chat interface, an API, or external channels such as Slack, Microsoft Teams, Telegram, Discord, Linear and WhatsApp, and can also generate lightweight dashboards, websites and internal tools. Organization-level features include shared company context, an org registry for reusing agents, tools and skills across teams, evaluation suites and traces for inspecting runs before and after shipping.

The original Buster product was an AI-powered analytics platform that used natural language to retrieve and present database information (an "AI data analyst"). The company pivoted to Keystroke and open-sourced the platform, with code published on GitHub under the keystrokehq organization.

Founding story

Buster was created by cofounders Dallin Bentley, an MISM graduate from South Jordan, Utah, and Blake Rouse, a strategic management student from Greer, South Carolina, together with software engineer Nate Kelley, after months of trial and error in BYU Marriott's Sandbox program, an 18-credit, two-semester entrepreneurship program run by the Rollins Center for Entrepreneurship and Technology. Rouse studied Product Management and Computer Science at BYU for three years before dropping out to start the company; he previously led product development at bootstrapped analytics startup DataSpark, which was acquired by Threecolts.

Business model

Software platform sold to companies for building and running internal agents and workflows, combining an open-source code-first framework with a hosted web application that manages deployment, integrations, credentials, permissions and logs.

Usage-based credits for platform consumption, with free credits for new accounts and a sales-assisted track for larger teams.

Traction

Y Combinator lists the company as active with a team size of 4. The platform is in open alpha and open source. The founders report internal production use of an agent built on the platform for recurring analytics and reporting tasks.

Latest developments

The company announced a pivot from Buster to Keystroke, open-sourcing its internal agents and automations platform and releasing it in open alpha with free trial credits. The team reported rebuilding the original Buster AI data analyst on Keystroke in about five minutes by pointing Cursor at the old Buster repository, producing an agent ("Delbert") connected to Postgres, PostHog, Metabase and Slack that answers ad-hoc questions, builds dashboards and sets its own triggers.