Herve
Hervé is a code review platform built for the agentic coding era, giving engineering teams a unified surface to review AI-generated pull requests. It captures AI agent sessions and attaches them to PRs, so reviewers see the full context—prompts, plans, and decisions—behind every change. The tool also prioritizes files by risk and flags plan deviations directly on the code, helping teams review faster without losing control.
- Artificial Intelligence
- AI Agents
- Developer Tools
- Software Only
Funding
Founders
Product
Problem
AI coding agents generate pull requests faster than human reviewers can meaningfully assess them. Reviewers are forced to read through large diffs without understanding the AI's intent, plan, or decision-making process, leading to missed risks, slower reviews, and a growing trust gap between human oversight and automated code generation.
Solution
Hervé provides a dedicated review surface for AI-generated pull requests, capturing the full agent session—prompts, plans, tool calls, and decisions—and attaching it to every PR. Reviewers see a change map that sizes files by modification volume and colors them by risk, a criticality score that highlights which files need attention, and in-place symbol navigation for an IDE-like experience without leaving the PR. The platform also tracks recurring patterns and team prompting habits, turning ad-hoc workflows into reusable skills. All code is read on demand and never stored; session data lives in a branch on the customer's own repository, ensuring IP never leaves their control.
Target Audience
Engineering teams actively using AI coding agents like Claude Code, Cursor, OpenAI Codex, or GitHub Copilot, who need a structured, context-rich review workflow for AI-generated pull requests.
Features
- Change map visualization that sizes files by lines changed and colors them by risk level
- Criticality scoring that flags the few risky files in a large PR, letting reviewers prioritize their time
- In-PR symbol navigation with definition and usage lookup, plus slash-command search, without checking out the branch
- Automatic capture of AI agent sessions (prompts, plans, tool calls) attached to each commit and PR
- Anti-drift detection that flags when the AI deviates from the original plan, with comments posted directly on the offending line
- Insights dashboard showing token consumption per file, skill usage, and recurring patterns across the team
- Git hook installation that backfills the last 30 days of local agent sessions for immediate data availability
- Architecture that stores code in memory only, never on disk, with session data kept in a branch on the customer's own repository