# Swarm > Swarm is an open-source, local-first multi-agent coding engine and daemon (`swarmd`) built for autonomous development, deterministic task execution, and worktree-isolated multi-agent collaboration. Swarm operates as a local background daemon (`swarmd`) communicating over private loopback Unix/HTTP sockets (`127.0.0.1:5555`) with interactive terminal (TUI) and desktop clients. It is designed from the ground up for software engineering workloads that require rigorous safety, reproducible tool use, and concurrent multi-agent coordination without risk of workspace corruption. ## Core Architectural Invariants - **Git Worktree Session Isolation**: Every agent session and execution runs in an isolated Git worktree branch (`.swarm/worktrees//`). Sibling agents never race against each other, overwrite shared files, or contaminate the host checkout during compilation or testing. - **Embedded Pebble LSM Key-Value Engine**: Headless `swarmd` embeds CockroachDB's high-performance Pebble KV engine directly in the Go binary. Zero external database dependencies (no PostgreSQL, Redis, or SQLite lock contention). Pebble powers sub-millisecond local state storage, event journals, and session persistence. - **Deterministic Task Program DAGs**: Multi-stage dependency graphs coordinate specialized agent roles separated by strict integration barriers. Specialized agents include: - `auto` / `plan`: Primary orchestration and goal decomposition. - `coder`: Autonomous code authoring and test authoring in isolated worktrees (deliberately restricted from raw shell/bash execution to prevent accidental host mutation). - `finder`: High-speed, read-only codebase discovery and semantic mapping. - `designer`: UI/UX layout and artifact design (HTML, Canvas, SVG, Tailwind). - `router`: Task brief hydration and model routing. - `compact`: Semantic context-window compaction. - **Tool Permission Boundaries**: Fine-grained tool governance with explicit human approval gates for filesystem mutation, command execution, and remote cloud infrastructure. ## Documentation Index ### Getting Started & Overview - [Overview](https://swarmagent.dev/docs): High-level introduction to Swarm, the daemon architecture, and client options. - [Why Swarm](https://swarmagent.dev/why-swarm): Architectural comparison with single-agent IDE extensions, detailing Git worktree isolation, Pebble KV, and multi-agent coordination. - [5-Minute Guide](https://swarmagent.dev/docs/5-minute-guide): Accelerated hands-on walkthrough to run your first task in Swarm. - [Quickstart & Installation](https://swarmagent.dev/docs/quickstart): Installing `swarm` and `swarmd`, connecting clients, and completing local onboarding. - [Host Your Swarm](https://swarmagent.dev/docs/host-your-swarm): Guidelines for deploying headless `swarmd` on dedicated Linux servers or cloud VMs. ### Multi-Agent Swarming & Task Programs - [Swarming Effectively](https://swarmagent.dev/docs/swarming): Core concepts for directing concurrent and staged agent swarms. - [Task Programs](https://swarmagent.dev/docs/swarming/task-programs): Authoring deterministic multi-agent DAGs with sequential stages, dependency barriers, and specialized roles. - [Iteration Swarms](https://swarmagent.dev/docs/swarming/iteration-swarms): Parallel exploration waves for rapid design alternatives, benchmark trials, and algorithmic options. - [Artifacts V3](https://swarmagent.dev/docs/swarming/artifacts): Native document, code, and UI artifact creation, revision tracking, and workspace materialization. - [Images & Visual Media](https://swarmagent.dev/docs/swarming/images): Generative image creation, asset remixing, and multimodal verification. - [Video Studio & Audio](https://swarmagent.dev/docs/swarming/media): Multi-clip timeline editing, HTML animation conversion, and soundtrack ingestion. ### Runtime, Access & Isolation - [Agent Profiles & Roles](https://swarmagent.dev/docs/agents): Configuring primary agents and subagents (`auto`, `coder`, `finder`, `designer`, `router`). - [Environments & Testbenches](https://swarmagent.dev/docs/environments): Ephemeral container and remote testbench leases for autonomous verification. - [Model Context Protocol (MCP)](https://swarmagent.dev/docs/mcp): Integrating external tool servers, resources, and custom prompts via MCP. - [Workers & Recurring Automation](https://swarmagent.dev/docs/workers): Worker V2 cron and interval scheduling for recurring codebase reviews and background audits. - [Permissions & Security](https://swarmagent.dev/docs/permissions): Granular approval policies, execution categories, and critical operation prompts. - [Swarm Security Architecture](https://swarmagent.dev/docs/swarm-security): Security invariants, secret privacy boundaries, and defense-in-depth isolation. - [Authentication & Remote Access](https://swarmagent.dev/docs/auth): Private token authentication and loopback broker configuration. ### Workspace & Storage - [Workspaces](https://swarmagent.dev/docs/workspaces): Multi-workspace cataloging, workspace maps, and root path boundaries. - [Worktrees](https://swarmagent.dev/docs/worktrees): Git worktree lifecycle management, promotion, lineage recall, and conflict resolution. - [Memory](https://swarmagent.dev/docs/memory): Account-level durable memory entries, operational rules, and project context. - [Themes](https://swarmagent.dev/docs/themes): Theme customization and palettes for Desktop and Web interfaces. - [Configuration Reference](https://swarmagent.dev/docs/configuration): Full specification for `swarm.conf`, CLI flags, and environment variables. ### Catalogs & Resources - [AI Model Catalog](https://swarmagent.dev/models): Directory of 734+ supported AI models across providers with context limits, pricing, and specs. - [0-to-1 Dedicated Machine Setup Guide](https://swarmagent.dev/resources/guides/zero-to-one-machine-setup): Complete engineering playbook for provisioning, hardening, and operating a dedicated Linux host for Swarm, embedded Pebble KV storage, cloud IMDS null-routing, and private loopback tunneling. - [Secure Production AI Agent Setup](https://swarmagent.dev/resources/guides/secure-ai-agent-setup): Practical guide to isolating autonomous AI agents, ambient authority minimization, package supply chain controls, and PR promotion pipelines. - [Benchmarks & Performance](https://swarmagent.dev/resources/benchmarks): Swarm task routing, planning accuracy, and token economics benchmarks. - [Full Technical Reference](https://swarmagent.dev/llms-full.txt): Complete, consolidated documentation manual for single-context AI ingestion.