Quickstart

Once CoDev is installed and configured (see Installation), this guide covers day-to-day usage — indexing your codebase with CodeGraph, launching the CLI and IDE extensions, switching models, and monitoring your usage.

Usage

CodeGraph

CoDev uses CodeGraph, a lightweight tool that builds a pre-indexed knowledge graph of your codebase. With it, the AI agents don't have to scan and grep your files on every read — which shortens response times and saves tokens.

To use CodeGraph, cd into the project you're working on and run this command once:

codevhub init

codevhub init builds the knowledge graph for the current project. CodeGraph then watches your codebase and auto-syncs whenever files change.

This creates a local .codegraph/ directory. You can commit it if you'd like to share the knowledge graph with your team.

Code files only

CodeGraph only works on programming-language files (e.g. .py, .js/.ts, .java, etc.); it does not work on unstructured files like documents (.md, .pdf, .docx, …). In other words, if your working folder contains only documents, CodeGraph won't help.

Start the CLI

Open a new terminal/PowerShell window.
cd into your project and type claude or opencode.
Trust the folder when prompted.
If asked whether to use an API key, choose Yes. Note: Claude Code defaults this to No (recommended) — you need to switch it to Yes.
Once you're in the CLI, you do not need to run /login. Test with a prompt like "hi"; if the model replies, you're connected and can start using it normally.

If Claude Code asks you to log in

If Claude Code forces a login when you prompt, it's because you didn't choose Yes above. Run codevhub config to reconfigure.

Language hallucination

A common issue with today's Chinese open-source models is language hallucination — they sometimes insert Chinese into the answer. To avoid this, say "use English only" explicitly in your prompt.

Select a model in the CLI

For OpenCode, if it isn't already set to MiniMax or GLM, type /model and choose MiniMax or GLM from the AI Gateway (only the first time — OpenCode remembers your choice afterward).
Claude Code and Codex don't let you pick an external model inside their CLI; they only let you choose a single model to replace Anthropic's and OpenAI's proprietary one.
To change the default model, run codevhub model and choose the model you want.

Extensions

Launch as usual inside your IDE (VS Code, JetBrains). If you've used it before, you may need to close and reopen the extension window.
Like the Claude Code CLI, the Claude Code extension doesn't let you pick a model — it uses the default model set during install. On first launch a sign-in screen may appear; wait a moment and it auto-bypasses into the chat window.
Continue lets you switch models flexibly right in the extension window. On first launch, wait a few minutes for Continue to load its config file.
The codevhub model command applies to the extensions too.

Switch between proprietary and self-hosted models

CoDev offers a flexible way to move between proprietary models and the self-hosted models.

To revert a tool to its original state from before CoDev was installed:

codevhub restore claude
codevhub restore codex
codevhub restore opencode
codevhub restore continue

# Restore all of the tools above
codevhub restore

To configure the tools to use the self-hosted models again, run codevhub config.

Restart your terminal/PowerShell after each switch. To resume your most recent CLI session:

claude -c
codex resume
opencode -c

codevhub login
codevhub logout

Upload logs

CoDev automatically collects and uploads session logs whenever a user runs any AI agent. Users can run codevhub upload to upload them manually.

Remove CoDev

Run codevhub remove to remove CoDev's configuration, then npm rm -g codev-ai to remove the package:

codevhub remove
npm rm -g codev-ai

Monitoring

You can visit the CoDev Analysis page (sign in with SSO) to see your own CoDev usage statistics. See the CoDev Analysis user guide for details.