Polyglav Documentation

Detailed reference for Polyglav. For the overview and features see the README, and for installation and usage see INSTALL.md.

• Agent fleets - single-purpose agent fleets, one scoped process per agent, deployment
• Agent swarms - swarm orchestration, delegation, types, auditor and team patterns
• Roles - role catalog, storage, delegation and permission rule
• API endpoints - HTTP JSON API for polyglav serve
• Architecture - agent core: the loop, engine, UI sinks, front-ends, extension points
• Commands & CLI - slash commands and headless CLI flags
• Configuration - config schema and keys
• Deployment - Docker deployment (image + Compose fleet)
• Eval harness - tool-use evaluation (polyglav eval), task fixtures, metrics
• Jobs - scheduled and durable jobs (polyglav jobs), cron scheduling, approvals
• Memory - bounded role, team, and job memory, automatic and manual
• Model Context Protocol - MCP client and server
• Plugins - bundled, layout, manifest, management
• Providers - providers, auto-detection, the chat event contract, adding a provider
• Security - permission model, threat model, data posture, prompt injection
• Sessions - session log format, message schema, compaction, persistence
• Skills - skill registry, storage, type injection
• Testing - running the mock test suite, per-file coverage map
• Tools - tool registry, bundled tools, tool policy, registration metadata
• Usage - step-by-step setup variations
• Workflows and usage - how work flows from a request to a result, and common patterns
• Writing tools for agents - how to design, name, and describe tools for the agent loop