Opensteer provides a platform for creating and running specialized AI agents that can browse the web, use tools, and execute scheduled tasks in the cloud or on local machines. The platform offers both hosted workspaces and an open-source local runtime, with agents defined through "harnesses" containing instructions, scripts, and state. Users can bring their own model subscriptions or use managed models, with pricing based on model and platform credits.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations and developers need AI agents that can perform real, ongoing work—browsing websites, interacting with tools, and running on schedules—but building and deploying such agents typically requires complex infrastructure and integration work. Existing solutions often lack flexibility in where agents run or how they are customized for specific workflows.
Solution
Opensteer provides a platform for creating and running specialized AI agents that operate in hosted cloud workspaces or on local machines. Agents can browse the web, interact with browsers, terminals, files, and third-party APIs, and run on schedules. The platform includes an open-source runtime that gives coding agents like Codex and Claude Code browser control and the ability to call project-specific code. Users define agents through "harnesses"—folders containing markdown skill instructions, deterministic scripts, and durable state—which can be customized for specific jobs. Opensteer Cloud enables agents to run 24/7 on managed servers with hosted workspaces, browser profiles, security controls, and team access.
Target Audience
Primary users are developers and teams building automated workflows, including those using coding agents like Codex and Claude Code, as well as organizations needing persistent, scheduled AI agents for browser-based tasks and tool interactions.
Features
- Open-source local runtime with CLI installation that attaches a browser and adds agent skills to coding agents like Codex, Claude Code, and Cursor
- Harness-based agent structure with separate folders for skills (markdown instructions), scripts (deterministic code), and state (SQLite, JSON snapshots, caches)
- Agents can call project-specific code directly, enabling them to inspect, run, and modify scripts as tools while working
- Cloud-hosted workspaces for 24/7 agent operation with schedules, browser profiles, security controls, and team access
- Template harnesses for common agent patterns that can be copied and customized
- Support for bringing your own Claude or ChatGPT subscription to run agents free, or using managed models with API pricing
- Browser and computer runtime billed at $0.20 per hour