--- title: Subagents description: "Specialized expert agents the main agent can call on demand" --- A subagent is a specialized "expert" that the main agent (Worker) can call on at any time during task execution. Each subagent has its own expertise, dedicated prompt, and can even run on a different model. In chat, simply type `@SubagentName` and it joins in to help—just like tagging a colleague in a group chat. --- ## Built-in Subagents Verdent ships with 6 subagents out of the box, covering the most common needs in software development: | Name | Default Model | What it's good at | |------|---------------|-------------------| | **General** | Follows main agent | A general-purpose research agent for complex problems, code search, and multi-step tasks—can help with almost anything | | **Fast Context** | Gemini 3 Flash | Quickly locating code and gathering contextual evidence; specializes in codebase indexing and is extremely fast | | **Verifier** | Sonnet 4.6 | Fast code review—after you make changes, tag it to catch obvious issues | | **Verdent Helper** | Haiku 4.5 | A Verdent usage assistant; ask it anything about how Verdent works | | **Multi-Model Planner** | Gemini 3.1 Pro; Opus 4.6; GPT-5.4 | Generates plans using multiple models (Claude, GPT, Gemini) and merges their outputs | | **Reviewer** | Gemini 3.1 Pro; Opus 4.6; GPT-5.4 | Full code review: logic errors, edge cases, hidden risks | Two subagents are special: **Multi-Model Planner** and **Reviewer** run 2–3 models in parallel. Each model thinks independently, produces its own output, and the results are then combined. All other subagents are single-model. --- ## How to Use Subagents In any conversation, type `@` to open the subagent selection menu (fuzzy search supported). Select the subagent you want and it will take over the current context. **Examples:** - After writing complex authentication logic: `@Reviewer please review this` → Reviewer checks it from security, logic, and edge-case perspectives. - For a tricky architectural decision: `@Multi-Model Planner` → multiple models propose solutions and you pick the best one. - After coding, quickly run `@Verifier` → faster than running full test suites for basic validation. --- ## Custom Subagents The built-in subagents cover most needs, but sometimes you want a specialist tailored to your project. You can create your own. ### Where to store them Custom subagents are stored as Markdown files in: ``` ~/.verdent/subagents/ ``` Up to 5 levels of subdirectories are supported for organization. ### File format Each subagent is a Markdown file with YAML frontmatter: ```markdown --- name: SQL-Expert description: Specializes in SQL optimization and database design, good at analyzing slow queries and indexing strategies color: blue model: claude-sonnet-4-6 --- You are a senior database expert focused on PostgreSQL performance optimization. When users provide SQL queries, you should: 1. Analyze the query execution plan 2. Identify performance bottlenecks (full table scans, missing indexes, etc.) 3. Provide concrete optimization suggestions and rewritten SQL ... ``` ### Key details - **Name:** must only include letters, numbers, and hyphens (e.g. `SQL-Expert`, `MyAgent123`). No spaces or non-ASCII characters. - **Color:** one of `dark-green`, `light-green`, `red-orange`, `amber`, `cyan`, `blue`, `pink`, `purple`, `verdent`—for easier visual recognition. - **Hot reload:** saving the `.md` file takes effect immediately—no need to restart Verdent. --- ## Managing Subagents in Settings All subagent management happens in **Settings → Subagents**. | Action | Description | |--------|-------------| | **View list** | All subagents are shown as cards. Built-in ones are labeled "Built-in"; plugin-based ones show plugin names. | | **Enable / Disable** | Toggle each subagent on or off without affecting others. | | **Create new** | Click **Add Subagent** and fill in Identifier (unique name), Model, Description, and Prompt. | | **Edit** | Hover a card to edit. Built-in subagents allow only Model and Description changes. | | **Duplicate** | Copy any custom subagent (duplicates are disabled by default until enabled). | | **Delete** | Custom subagents can be removed (with confirmation). Built-in ones cannot be deleted. | ### Model selection differences - **Multi-Model Planner** and **Reviewer** support multi-model selection (2–3 models at once). - Other subagents use single-model selection. - **General** has a special option: it can follow the main agent's model automatically, avoiding separate configuration maintenance. --- ## Use Cases ### Code review After implementing complex logic, tag `@Reviewer`. It checks correctness, edge cases, and potential bugs—saving you from manual line-by-line inspection. ### Multi-perspective architecture design For major decisions (e.g. microservices adoption, database choice), use `@Multi-Model Planner` to get independent proposals from multiple top models and reduce decision risk. ### Fast validation After code changes, `@Verifier` quickly checks syntax, types, and basic issues—faster than full test suites, more reliable than no checks at all. ### Custom domain experts Create subagents tailored to your project: - SQL optimization expert aware of your schema and query patterns - Frontend code standards guardian enforcing your ESLint and design system rules - Documentation assistant that adapts technical content into clear, reader-friendly writing --- ## Notes - Custom subagents must be configured with a valid model to be enabled. If you use BYOK and the key expires or the model is retired, the subagent will be disabled automatically. - Built-in subagent models can be changed, but the defaults are tuned—modifying them may reduce performance. - **Description matters:** it is injected into the main agent's context and influences when subagents are called. Better descriptions lead to smarter routing. - Subagents cannot call each other. They can only be invoked by the main agent (Worker)—no deeper nesting allowed.