Contents
Developers
Building and self-hosting
Install, build, connect, and self-host. Start with Install, or jump to the reference for manifests, step config, and tool filtering.
Get started
- InstallInstall the macOS app or self-host on Linux, then complete first-launch setup.
- Your first workflowThe first workflow should be small enough to review but real enough to produce evidence. A recurring research brief,...
- Core conceptsThese docs describe platform subsystems that cut across workflow and multi-agent mode.
Goals & Auto-improve
- Auto-improve and reportingStart here. This is the map of how a workflow keeps itself healthy and moving toward its goals, and how that work is...
- Auto-improve frameworkBounded self-improvement: run evidence, maintenance, strategic review, and audit trail.
- Human approvals & feedbackHow workflows ask humans for input: approvals, choices, and blocking questions.
- Learning & knowledgeHow runs turn into reusable skills, learnings, and knowledge for future runs.
Crew & memory
- Chat connectorsA platform-agnostic bot framework that allows users to interact with the agent system from messaging platforms...
- SkillsSkills are reusable instruction sets that guide workflow agents on how to handle specific tasks. They provide domain...
- Agent memoryPersistent memory for the multi-agent chat mode. Memories survive across sessions and are stored as markdown files...
- Crews & multi-agentThese docs cover delegation, multi-agent chat, and coordination between manager and worker agents.
Workflows
- Workflows overviewThese docs cover workflow design, execution, and workflow-scoped runtime behavior.
- SchedulesCron and cadence schedules, quiet hours, and per-run timing rules.
- MonitoringRun evidence, status, and health: what to look at when a workflow misbehaves.
- Cost & logsHow token usage, run cost, and logs are measured and stored.
- Browser automationBrowser workflows use the agent-browser skill and the managed agent_browser tool. Set capabilities.browser_mode to...
- Model tiersTiered allocation is the workflow's auto-selection mode for runtime LLM choice.
Connect
- AI plans & model routingThis document outlines the system for managing LLM configurations, explicit model selection, and retries on the...
- MCP bridgeThe MCP Bridge Layer is a critical architectural component of AgentWorks. It acts as a universal proxy and...
- BrowserAgentWorks uses the managed agent_browser tool for all browser automation. The browser can run headlessly in the...
- SecretsSecrets allow sensitive credentials (API keys, database passwords, tokens) to be securely injected into agent...
Reference
- Workflow manifestThe workflow.json format: goals, schedules, steps, and how the backend discovers them.
- Step config formatThe step_config.json file stores step-specific agent configurations (LLM models, tool selections, execution...
- Tool filteringThe Tool Filtering and Configuration System provides a powerful, multi-layered mechanism to control exactly which...