Propel builds AI agents that review pull requests at the level of senior engineers.
The primary bottleneck in software engineering is shifting from creation to verification. The new challenge lies in the ability of software engineers to perform fast, high-quality reviews of the code being deployed to production.
High quality code review requires a detailed understanding of code changes in the full context of the codebase and its dependencies.
This must work for new repositories - but also large enterprise codebases like the ones that Tony Dong, CEO, saw when he was VP of Engineering at Rippling. Some key challenges:
- Best practices and patterns are spread across large repositories
- Related logic can exist far from the changed code
- Critical behavior is defined in third-party dependencies
Propel's AI agent must:
- Search first-party code and third-party dependencies
- Search with semantic understanding and precise symbol-level matching
- Search quickly to provide near-real-time feedback to engineers in the review loop
- Search efficiently to drive down token cost and LLM latency
Propel chose Chroma to power agent search.#
Customer repositories are continuously indexed in Chroma Cloud, enabling the agent to retrieve relevant context across the entire codebase, even when related logic is far from the files under review.
Semantic retrieval is complemented with regex‑powered search and AST‑based analysis, allowing the agent to locate exact symbols, structural patterns, and syntactic intent. This grounds recommendations in concrete code and helps validate them against real usage patterns.
Chroma Cloud provides strong isolation between customer codebases and secure access controls, enabling Propel to confidently serve enterprise customers with strict security requirements.
The result is an agent that can reason across large codebases while grounding it's feedback in precise symbol-level evidence.
Chroma Cloud plays a core role in enabling Propel's agent to reason more like an experienced engineer, drawing on knowledge embedded across both the codebase and its open-source dependencies.
Agents start with context. Getting the right context requires search. Chroma makes indexing easy with collection forking and provides powerful search semantic, regex, and full-text making it ideal for code search applications.
Want to learn more about how to implement code search?
Check out our series on Chroma For Code.
