AgentWorksProduct / the operating loop
The agent stops. The workflow should keep improving.
AgentWorks runs recurring business workflows across sales, marketing, support, finance, and operations. It connects the agents and tools you already use, keeps every run inspectable, and turns evidence plus human judgment into a better next run.
Define the outcome, not only the prompt
Give recurring work a goal, owner, schedule, output contract, and approval boundary. Route each step across Claude Code, Codex CLI, Cursor, Gemini, browser sessions, MCP tools, APIs, and lower-cost models.
Keep the operating record
Every run retains reports, logs, screenshots, source health, cost, approvals, and goal evidence. Pulse separates bugs from goal drift so operators review exceptions instead of every run.
Improve with human judgment
Goal Advisor critiques the workflow and asks bounded questions. Auto Improve applies low-risk fixes, proposes larger changes, and turns approved knowledge into better context for the next run.
What is AgentWorks?
AgentWorks is an open-source desktop control plane for AI and coding-agent fleets. It wraps Claude Code, Codex CLI, Cursor, Gemini, browser workers, MCP tools, schedules, secrets, Pulse, costs, approvals, evidence, reports, and reusable skills into one operating loop.
How does AgentWorks keep every run inspectable?
Every run retains reports, logs, screenshots, source health, cost, approvals, and goal evidence. Pulse separates bugs from goal drift so operators review exceptions instead of every run.
What do Pulse and Goal Advisor do?
Goal Advisor critiques the workflow and asks bounded questions. Auto Improve applies low-risk fixes, proposes larger changes, and turns approved knowledge into better context for the next run.
What is Auto Improve, and who approves changes?
Auto Improve proposes bounded workflow changes from evidence; humans approve risky writes, production changes, major replans, and promoted skills.