A search engine that writes code: it’s not science fiction but the latest move by Perplexity, the San Francisco startup known for its conversational approach to search. According to Business Insider, the company has built an internal coding tool code-named “Teammate,” already used by its engineers since May, and is evaluating a public launch.

The news places Perplexity in a crowded arena — from GitHub Copilot to Cursor to Anthropic’s Claude Code — but with an atypical background. While other players come from development platforms or generalist large language models, Perplexity hails from real-time search and retrieval-augmented generation (RAG). This starting point shapes how Teammate might interact with codebases and documentation.

A coding assistant rooted in search brings two promises and a potential conflict. The first promise: the ability to tap into updated external sources — official documentation, public repositories, Stack Overflow threads — with the same fluidity Perplexity uses to answer current affairs questions. The second: integration with a company’s internal knowledge base, turning the tool into a search engine for its own code assets. For teams managing huge monorepos or poorly documented legacy systems, this blending of coding and information retrieval could reduce daily friction more sharply than purely predictive completion.

But the second frontier holds the larger friction. A company that lets an interrogable LLM index its repositories — with access logic, secrets, configurations — exposes itself to data sovereignty risks that go well beyond chat privacy. If the tool is cloud-hosted, each query becomes a potential vector for intellectual property leakage, unless strict safeguards and explicit service agreements are in place. This is structural: the industry’s default of pushing everything into the cloud clashes with the need to keep code (and the context that explains it) under on-premise or self-hosted control.

It’s not just a compliance checklist item. The stakes touch the core competitive advantage of any tech enterprise. As SaaS giants embed code assistants in their IDEs, the demand grows for solutions that run on one’s own infrastructure, with verifiable guarantees and no dependency on external APIs. In that light, a publicly available Teammate would have to tackle the deployment model question from day one. Being faster than Copilot at suggesting a function won’t be enough; it will need to convince those developing proprietary software that every processed token leaves no trace on third-party servers.

Perplexity built its reputation on source transparency and real-time synthesis. Applying the same philosophy to code means accepting a radically higher level of trust. For those already evaluating AI coding tools, the debate shifts from completion quality to data residency and long-term TCO. While we wait for an eventual public debut, Teammate’s emergence already serves as a reminder: the line between search and development is getting thinner, and crossing it without a local-control strategy could prove expensive.