> ## Documentation Index
> Fetch the complete documentation index at: https://personal-ce79cb71.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Normal vs. Deep Review Mode in BugViper

> Compare BugViper's Normal and Deep review modes. Normal is fast with one generalist agent; Deep runs three specialized agents in parallel for thorough analysis.

BugViper offers two review modes that give you control over the trade-off between speed and depth. Both modes build a full call graph, score files by blast radius, and run a verifier pass — the difference is in how many agents review your code and how thoroughly each batch is analyzed. You can set a default for your repository and override it at any time using a comment on the PR.

<Tabs>
  <Tab title="Normal Mode">
    Normal mode routes all file batches through a single **generalist agent** that understands bugs, security issues, and performance concerns in one pass.

    **How it works:**

    * Your PR's changed files are grouped into up to **4 batches** using Louvain community detection.
    * Each batch is reviewed sequentially by one generalist agent.
    * The agent has access to the call graph, blast radius scores, and the PR diff for full context.
    * Findings go through the verifier pass and dedup before being posted.

    **Best for:**

    * Everyday pull requests with moderate scope
    * Fast feedback loops where speed matters
    * Smaller PRs where a single agent can hold all the context
    * Teams that want consistent, low-latency reviews on every commit

    Normal mode is the default configuration for new repositories.
  </Tab>

  <Tab title="Deep Mode">
    Deep mode replaces the single generalist with **three specialized sub-agents** that run in parallel, each inside its own isolated E2B sandbox. Every batch is examined from three distinct angles simultaneously.

    **The three sub-agents:**

    | Agent               | Focus                                                                                     |
    | ------------------- | ----------------------------------------------------------------------------------------- |
    | **Bug Correctness** | Logic errors, incorrect assumptions, edge cases, off-by-one errors, null handling         |
    | **Security**        | Injection risks, authentication flaws, insecure data handling, dependency vulnerabilities |
    | **Performance**     | Inefficient algorithms, unnecessary allocations, blocking I/O, N+1 query patterns         |

    **How it works:**

    * Your PR's changed files are grouped into **2 batches**.
    * Each batch is sent to all 3 sub-agents simultaneously in parallel E2B containers.
    * All findings are collected, verified against the diff, and deduped before posting.

    **Best for:**

    * Critical infrastructure changes
    * Security-sensitive code (auth, payments, data access)
    * Performance-critical paths
    * Large refactors where a single agent might miss cross-cutting concerns
    * Pre-release or pre-merge reviews on important branches
  </Tab>
</Tabs>

## Mode Comparison

| Feature               | Normal                          | Deep                                                |
| --------------------- | ------------------------------- | --------------------------------------------------- |
| **Agents**            | 1 generalist                    | 3 specialists (Bug, Security, Performance)          |
| **Batches**           | 4                               | 2                                                   |
| **Speed**             | Fast                            | Slower (parallel agents offset cost)                |
| **Depth**             | Broad coverage in a single pass | Deeper per-domain analysis                          |
| **Sandbox isolation** | Single E2B sandbox              | One E2B sandbox per sub-agent per batch             |
| **Use case**          | Everyday PRs, fast feedback     | Critical changes, security reviews, large refactors |

## How to Trigger Each Mode

BugViper uses the review mode configured for your repository by default. You can change your default mode from the BugViper dashboard under repository settings.

To trigger a review from a PR comment, use:

```
@bugviper full review
```

This always uses your configured default mode. The `@bugviper review` command also respects the same default and is useful for incremental reviews as you push new commits.

<Tip>
  **Not sure which to use?** Start with Normal mode for most PRs — it catches the majority of real issues with minimal latency. Switch to Deep mode when you're merging into a protected branch, touching authentication or payment logic, or making a change that affects a high blast-radius module. You can always re-run in Deep mode by commenting again after a Normal review.
</Tip>