Keystroke

Keystroke is a platform for teams that want to run real AI agents inside their own company without assembling a framework, a job queue, a sandbox provider and an observability stack themselves. Agents are authored in the product or deployed from your own repository, and can equally be built with a coding agent such as Cursor or Claude Code.

What an agent gets

Every Keystroke agent is given persistent memory, a built-in workspace with a filesystem and a bash shell, custom tools, and access to skills and project files. Connectivity is deliberately broad: over 1,000 app integrations, any HTTP API, and any MCP server. Once built, an agent can be used from Keystroke's own chat, from external channels including Slack, Microsoft Teams, Telegram, Discord, Linear and WhatsApp, or called directly through an API.

Systems, not just chatbots

The more distinctive half of the product is what Keystroke calls systems — durable workflows that combine agent steps, plain action steps, LLM steps and human approval hooks. Workflows use typed input and output (Zod), support durable sleeps and step replay so a long-running process survives restarts, and can be started from a schedule, a webhook or an event in a connected app such as a merged pull request or a successful Stripe charge. Agents can also spin up lightweight dashboards and internal tools as artifacts.

Testing and inspection

Because these systems touch production data, Keystroke ships evaluation suites that run against an agent before changes go live, plus per-run traces covering the chat, the timeline, the tools called and the human decisions taken.

Pricing

The Hobby tier is free forever and includes agent and workflow building, deployment from your own repo, triggers and webhooks, MCP integrations, bring-your-own-keys and run history, with $1 of monthly usage credit, one project, 5 GB of file storage and a 10-minute agent timeout. Pro is $20 per month and includes $20 of usage credit, five projects, 50 GB of storage, higher concurrency and 30-minute timeouts. An Organization tier adds role-based access control, agent guardrails, spend limits and unlimited concurrency.

Usage is metered rather than seat-based: $0.01 per agent run, $0.01 per workflow run, $0.005 per empty poll trigger, sandbox compute at roughly $0.067 per hour charged at cost, and Exa-powered web search and fetch at $0.007 and $0.001 respectively. Models routed through Keystroke's AI gateway carry a 1.1x markup on the provider's list price, or you can bring your own keys at no markup.

Who it is for

Keystroke is aimed at internal platform, operations and GTM teams who need agents that hold context across a company, act with approval where it matters, and can be tested and audited like any other production system.