guilds.run is an open-source multi-agent orchestration platform that runs on local machines, providing full run observability, persistent memory, sandboxed compute workspaces, and configurable connectors across heterogeneous language models.
What the product does and how it is positioned
guilds.run is an open-source system designed for deploying and managing teams of collaborative AI agents on user-controlled infrastructure. The platform addresses opacity in multi-agent execution by exposing the exact prompts submitted to models, tool execution outputs, and every memory modification across runs.
Each agent operates within an isolated workspace equipped with a terminal, script storage, and an optional persistent browser. Organizations can coordinate agents across distinct departments, attach scoped third-party connectors, and utilize the built-in Studio coach to diagnose workflow issues and refine agent instructions.
Source-supported ways to use the product
Teams of agents scan social timelines, score incoming posts against reply criteria, draft responses matching specific guidelines, and queue entries for operator review.
Configured agent recipes read shared support inboxes, tag and route issues, and prepare initial draft answers for human agent approval.
Automated agents inspect code repositories to organize issues, review incoming pull requests, and maintain development backlogs.
Agents research external prospect profiles, write introductory outreach communications, and log interaction threads into structured database tables.
The documented workflow, where available
Clone the open-source repository and deploy the local services using Docker Compose.
Define agents, standing task instructions, schedules, and shared database schemas directly or apply a preconfigured team recipe.
Provide model credentials from chosen providers and assign third-party service connectors to the organization, guild, or individual agents.
Inspect live runs, review tool calls and memory changes, and adopt prompt corrections suggested by the Studio coach.
Operating autonomous agent teams often presents debugging hurdles when multi-step outputs diverge from expectations. guilds.run addresses this by capturing comprehensive execution records for every run, exposing the verbatim prompts delivered to models, the arguments and return values of every tool call, and the resulting state changes in memory.
Building on this audit trail, the Studio feature functions as an automated diagnostic tool. It analyzes execution histories, shared notes, and memory tables to detect recurring failures, such as agents writing text exceeding length constraints. Studio drafts exact diffs for agent instructions, allowing operators to verify and publish adjustments before any live behavior changes.
Agents in guilds.run are not limited to stateless API calls; each agent has access to an isolated computing environment. Workspaces include dedicated terminals, reusable script execution capabilities, and browser sessions that remain logged in between scheduled runs to facilitate continuous web-based workflows.
Integration management supports broad model provider flexibility alongside granular tool permissions. Agents within the same guild can run differing models, ranging from commercial frontier APIs to local endpoints. Connectors to services such as GitHub, Notion, Gmail, and on-chain tools can be scoped strictly to the agents requiring them.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
guilds.run is open-source software that runs locally on the user's machine or on self-hosted servers using Docker Compose.
The platform supports frontier models including Claude, GPT, Gemini, and Grok, as well as open-weight models such as Qwen, DeepSeek, Kimi, Llama, Mistral, and GLM via local endpoints or provider keys.
The system records every execution step, showing the exact prompt seen by the model, each tool call and its output, and all persistent memory changes made across runs.
Each agent workspace includes a terminal, reusable script storage, and an optional web browser that retains login state across executions.
Connectors are managed with granular access controls, allowing administrators to attach an account to an entire organization, a specific guild, or an individual agent.
