Skip to main content
Docbot

​
Introduction to Docbot

Docbot is a CLI agent that helps you keep documentation up to date. It reads your docs and codebase, proposes a concrete plan with file-level operations, and only writes changes after you approve.

​
Key features

Codebase-aware documentation

Finds gaps and stale pages by reading your code, not vibes. Docbot analyzes your actual implementation to suggest meaningful doc improvements.

Semantic search that works

Combines semantic and exact match search with reranking to find exactly what you need in your docs and code.

Interactive planning

You approve the plan before anything touches your files. See exactly what changes will be made before they happen.

MDX-first output

Structured edits instead of giant blob rewrites. Docbot understands MDX components and makes targeted changes.

​
How it works

Docbot follows a structured workflow to improve your documentation:
1

Analysis

Scan docs and codebase to find gaps, duplicates, and stale content
2

Planning

Propose a structured set of operations: create, update, move, delete, or consolidate pages
3

Execution

Apply changes with MDX edits that are component-aware when relevant
4

Review

Verify the changes and re-scan for obvious misses

​
What makes Docbot opinionated

Docbot is designed with specific tools and frameworks in mind to enable rapid development:
  • Docs frameworks: Targets MDX-based doc sites and automatically detects Mintlify project structure (via docs.json). Support for other frameworks like Fumadocs, Nextra, and Docusaurus is planned.
  • Vector store: Uses Qdrant (required) for embeddings. It’s easy to run locally and performs well.
  • Models/provider: Currently uses Vercel AI Gateway via AI_GATEWAY_API_KEY. You can configure models in the config file, and support for other providers is planned.
  • Runtime: Requires Bun. This will not change.
While Docbot is opinionated today, it’s not meant to stay tied to a single docs framework or provider forever. The architecture allows for expansion as needs grow.

​
Requirements

Before using Docbot, make sure you have:
  • Bun runtime installed
  • Qdrant vector database (local via Docker or remote)
  • rg (ripgrep) for fast exact-match search
  • AI_GATEWAY_API_KEY environment variable set (Vercel AI Gateway)
The docbot init command will help you set up a local Qdrant instance via Docker automatically.

​
Performance notes

Docbot is under-optimized today but still provides significant time savings:
  • Indexing speed: Running a full cycle across multiple pages may take 5–10 minutes
  • Still worth it: Much faster than the hours you’d spend updating docs manually
  • Future improvements: The overall flow will continue to improve, reducing both time and token costs

​
Next steps

Quick start

Get from zero to running your first Docbot task in minutes

Installation

Detailed installation instructions and environment setup

Build docs developers (and LLMs) love

Get started for freeTalk to us