Curing AI Code Hallucinations: GitHits' Plan to Fix Coding Agents

WILMINGTON, DE – June 16, 2026 – The age of AI-assisted software development is a study in contradiction. Developers, now armed with powerful coding agents like GitHub Copilot and Claude Code, can generate functions and applications at unprecedented speeds. Yet, this new velocity comes at a cost: a rising tide of errors, security flaws, and nonsensical “hallucinations” that can turn a time-saving tool into a time-consuming liability. This growing “verification debt,” where the speed of AI generation outpaces human review, is the critical problem a new startup, GitHits, aims to solve. The company announced today it has raised $1.75 million in pre-seed funding to build what it calls the “Google for Code,” a specialized search engine designed to ground AI agents in reality.

The funding round, which saw participation from Vendep Capital, Trind, and prominent angel investors like LlamaIndex co-founder Jerry Liu, signals a significant bet on a new category of AI infrastructure: tools that don't generate code, but make the code generators smarter, safer, and more reliable.

The 'Hallucination Headache' in Modern Development

For any developer who has spent hours debugging AI-generated code, the problem is intimately familiar. AI models, trained on vast but static snapshots of the internet, often produce code that looks correct but fails in practice. A recent Stanford-led study found that up to 42% of AI-generated code snippets contained these kinds of hallucinations. The consequences are tangible, with one industry report finding that AI-authored code contains 1.7 times more issues on average than code written solely by humans.

These are not minor typos. The errors range from logic flaws and performance bottlenecks to critical security vulnerabilities, which some studies suggest are present in nearly half of all AI-generated code. One of the most pernicious issues is the “phantom dependency,” where an AI confidently suggests using a software library or package that simply doesn’t exist. This not only sends developers on a wild goose chase but also opens the door for malicious actors to create and upload malware under these hallucinated names.

“Coding agents are great at navigating your local codebase,” explains GitHits CTO Olli-Pekka Heinisuo. “The problem is that modern software doesn’t stop at the repository boundary. A large part of the system lives in frameworks, libraries, SDKs, and other open-source dependencies. Agents can’t inspect those nearly as well, so AI has to guess, and it produces code that looks correct but doesn’t work in practice.”

This gap forces developers into a frustrating loop of trial-and-error, where debugging the AI’s output can take longer than writing the original code manually. As one senior developer at a cloud infrastructure company noted anonymously, “The AI gets you 90% of the way there, but that last 10% is a minefield. You're trying to figure out why a function it invented doesn't work, and you end up questioning your own sanity.”

Building the 'Google for Code'

GitHits is positioning itself not as a competitor to the large language models from OpenAI, Anthropic, and Google, but as an essential complement. “GitHits doesn’t compete with Codex, Claude Code, or Cursor, but complements them by bringing open-source code as context for agents to end retry loops and reduce token consumption,” says Heinisuo.

The company’s solution is an AI-native, version-aware index of all public open-source code. Instead of guessing, a coding agent integrated with GitHits can query this index to find working examples of how a specific library function is implemented, inspect the source code of its dependencies, and check for known vulnerabilities. It provides the real-world context that models, by their nature, lack.

“Our vision is to index all public open-source code,” states CEO Jaakko Timonen. “With this funding, we are launching the beta version of the product today, and the first commercial version later this year.”

The initial product is a command-line interface (CLI) tool, launched today on the popular tech site Product Hunt, which gives AI coding agents a set of tools for finding working examples, navigating dependency sources, and inspecting software packages. By building a version-aware index on demand, GitHits ensures that the information is current and relevant to the specific versions of libraries a project is using, tackling the problem of outdated training data head-on.

A Niche Bet in a Crowded Field

The market for AI tools is intensely competitive, with billions flowing into both foundational models and general-purpose AI search platforms. US-based Exa, for example, recently raised a massive Series C to build a general search engine for AI agents. GitHits is deliberately taking a different path, focusing its efforts exclusively on the complex, structured world of source code.

Heinisuo clarifies the distinction: “Exa is building a general-purpose search for AI. GitHits focuses only on code.” This niche focus is precisely what attracted investors. They see a massive, underserved market in providing specialized, domain-specific intelligence that makes the entire AI ecosystem more effective. Rather than building another model, GitHits is building the vital connective tissue that allows existing models to function reliably in a professional software engineering context.

“We'd been watching GitHits since it was just an idea, and what convinced us was the team that formed around it,” says Timo Felin, Partner at Vendep Capital. “Olli-Pekka is a quiet legend in open source and has lived inside this problem for years. At this stage, you invest in people, and this was an easy call.”

From Martian Skies to the Modern Codebase

The investor's confidence is rooted in the team’s deep expertise, particularly that of its CTO. Heinisuo is a veteran of the open-source world, best known for developing and maintaining opencv-python, a critical software package for computer vision with over 100 million downloads. In a testament to its reliability, the package was even used by NASA in its Ingenuity helicopter, which successfully flew on Mars.

The idea for GitHits was born directly from Heinisuo's years of experience. While working at the AI consulting firm Softlandia, he grew frustrated by repeatedly giving colleagues the same manual search techniques to solve problems related to open-source code. He realized the process could be automated and scaled with AI, solving a pain point he had “lived inside…for years.”

He shared the idea with his colleague Jaakko Timonen, who became the company's CEO. Together, they assembled a team of four co-founders and spun the new company out with Softlandia’s support, turning a long-standing frustration into a funded venture. This origin story—a solution emerging from a real, persistent problem identified by a domain expert—provides a compelling narrative for how innovation truly moves forward: not just with breakthrough algorithms, but with practical tools that solve the day-to-day challenges of the people building the future.