Aide
Aide is a developer tools lab that builds AI agents for software engineering, using inference-time scaling to improve how code is written, debugged, and maintained. Its agent framework, built on a minimal toolset and LLM-assigned reward scaling, achieves state-of-the-art results on the SWE-bench Verified benchmark, resolving 62.2% of issues. The system runs up to five parallel trajectories per problem to explore the solution space.
- Developer Tools
- AI Agents
- Artificial Intelligence
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Traditional software engineering benchmarks and development workflows often rely on rigid, pre-defined frameworks that constrain language models rather than letting them explore solutions freely. This limits the effectiveness of AI coding agents, particularly on complex, real-world GitHub issues that require multi-step debugging and iterative problem-solving.
Solution
Aide provides a developer tools lab focused on reshaping how software is written, debugged, and maintained through inference-time scaling and agentic systems. The company's approach inverts conventional agent design by giving the language model a minimal toolset—file listing, opening, searching, editing, and terminal access—and using a reward-scaling framework to guide its trajectory. Each agent step generates an observation and an alternative action, enabling exploration of promising approaches without committing to a single path. By running multiple trajectories (up to five) with substantial compute, Aide's system can exhaust the solution space available to the model. This methodology has achieved state-of-the-art results on SWE-bench Verified, resolving 62.2% of issues, demonstrating that scaled test-time compute beats over-engineered frameworks.
Target Audience
Primary users are software engineers, AI researchers, and developer tooling teams seeking advanced automated coding assistance for complex debugging and maintenance tasks, particularly those working with GitHub issue resolution in production codebases.
Features
- State-of-the-art 62.2% resolution rate on SWE-bench Verified, improving baseline Sonnet 3.5 performance from 26%
- Minimal agent toolset: List Files, Open File, Str_replace_editor, Attempt Completion, RipGrep search, and Terminal Access
- LLM-assigned reward scaling per tool use step, inspired by SWE-search team's approach, rating how effective each action was toward solving the issue
- After each step, the agent generates both an observation and an alternative action to enable flexible trajectory exploration
- Mean trajectory reward calculation that is consistent across trajectory lengths, allowing equal preference for both long (≥20 steps) and short (≤10 steps) solution paths
- Test-time scaling framework that runs up to 5 trajectories per problem, exploiting LLM non-determinism to expand solution space coverage