An Agent is how Dosu shows up on a platform like GitHub, Slack, or Microsoft Teams to answer questions and take action. This page explains what an Agent is, what you can configure, and how it connects to the rest of Dosu.
What it is#
An Agent connects Dosu to a specific place where your team works whether a GitHub repository, a Slack channel, or a Microsoft Teams channel- and lets Dosu respond there. When someone asks a question, files an issue, or opens a pull request, the Agent draws on the Libraries it is connected to and replies in context.
Each Agent has its own configuration: the platform it runs on, the Libraries it reads from, when and how it replies, and the guidelines it follows. One organization can run many Agents at once, for example a low-noise Agent on a public repository and a more active one in an internal support channel.
Agents were previously called Deployments.
Why it matters#
For a growing engineering organization, knowledge is scattered across many repositories, channels, and teams, and it walks out the door when people leave. Agents put answers where the questions already happen instead of asking people to go find documentation.
- A new contributor opening an issue gets pointed at the relevant code, prior discussions, and Documents without a maintainer stepping in.
- A teammate asking in Slack gets an answer grounded in the connected Libraries rather than a half-remembered thread.
- Behavior is configurable per Agent, so a public repository, an internal channel, and a customer-facing space can each follow their own rules.
Dosu scales from a single repository to an entire company.
How it works#
You create and manage Agents on the Agents page. Only owners and admins can create or change them. To add one, open create an Agent, pick the platform and the repository or channel, choose a Library, and set the reply behavior.
Connected Libraries. An Agent can read from multiple Libraries. You configure which Libraries an Agent searches through the Agent's library membership settings. When connected to multiple Libraries, the Agent automatically scouts all of them for relevant information, then performs deep research in the most promising ones using a meta-orchestrator that coordinates across Libraries. Sources and citations from all connected Libraries appear in responses, though the Agent only returns information from Libraries the requesting user can access. Selecting Libraries connects the Agent to all of each Library's Sources at once.
Reply modes. Reply mode controls when an Agent posts in a channel or thread.
- Mention (default): Dosu replies only when explicitly mentioned with
@dosu. This is the lowest-noise option and a good starting point. - Auto Draft: Dosu writes a response automatically and saves it as a draft on the Review page instead of posting. A person edits, approves, or declines it before it goes out. A direct
@dosumention still posts immediately. - Auto Reply: Dosu posts responses automatically. Before publishing, Dosu evaluates the response, and replies that are speculative, generic, or aimed at a specific person or team may be held for Review anyway.
Response guidelines. Each Agent can carry free-text guidelines for tone, formatting, and scope, for example, narrowing which topics it should answer. These guidelines shape replies and also factor into whether an Auto Reply response posts or waits for review. See Custom Guidelines.
Platform behavior. What an Agent does depends on where it runs.
- GitHub: respond on issues, pull requests, and discussions. See GitHub.
- Slack and Microsoft Teams: answer in channels and direct messages. Direct messages always reply automatically and let each person pick which Library to use. These platforms are part of the Teams plan. See Slack and Microsoft Teams.
Staying in control. You do not have to let an Agent post unsupervised. Start in Mention or Auto Draft mode, watch what Dosu produces on the Review page, and move toward Auto Reply once you trust it. An Agent can be paused to switch it off without deleting it, then resumed later.
Multi-Library Agents#
An Agent's library membership determines which Libraries it searches. A single-library Agent behaves exactly as before: it searches only that Library's Sources. An Agent connected to multiple Libraries automatically searches all of them and synthesizes answers from the most relevant ones.
How it works. When you ask a question, the Agent first scouts every connected Library for relevant material, then performs deep research in the Libraries that look promising. The Agent decides which Libraries are relevant for each query using a meta-orchestrator, so you don't configure a fixed set of Libraries per topic. Sources and citations from all connected Libraries appear in responses.
Access control. The Agent only returns information from Libraries the requesting user can access. An Agent connected to three Libraries will only search the subset of those Libraries visible to the person asking.
Configuring library membership. You configure which Libraries an Agent searches through the Agent's library membership settings on the Agent detail page. The Library where the Agent's channel or repository lives is always included. You can add or remove other Libraries from the organization. When you change an Agent's library membership, its search scope changes immediately.