Heso

Heso develops specialized AI agents that handle routine, well-defined tasks more efficiently than large frontier models, allowing the primary model to focus on planning and judgment. Their first product, Soru-1, is a coding sub-agent that integrates into existing developer tools, executing code-related work faster and reporting results back to the main model. This approach reduces latency and cost while preserving the capabilities of the larger AI system.

Traction –/10

San Francisco, United States · HQ

110+ followers

AI Agents
Artificial Intelligence
Developer Tools
Software Only

Updated 10 days ago

Funding

Funding rounds are not available yet.

Founders

Founder details are not available yet.

Product

Problem

Frontier AI models are highly capable at reasoning through complex problems, but they are often used for routine, well-defined tasks like reading files, searching code, making edits, and running tests. Running these routine tasks on the most capable models is slow and costly, and it diverts the model's attention from the high-level planning and judgment that only it can perform.

Solution

Heso builds specialist AI agents designed to take on these defined, routine jobs from a frontier model, execute them much faster, and report the results back. This allows the frontier model to remain in charge of the overall plan and judgment, while the specialist handles the execution. Their first product, Soru-1, is a coding sub-agent that works inside existing developer tools, taking on coding tasks and reporting back to the primary model. This division of labor reduces latency and cost while preserving the capabilities of the larger AI system, ensuring the right model is applied to each part of the job.

Target Audience

Primary users are developers and engineering teams who use frontier AI models for coding tasks and want to reduce latency and cost by offloading routine work to specialized sub-agents.

Features

Soru-1 is a specialized coding sub-agent that integrates into existing developer tools.
The agent takes defined jobs from a frontier model, executes them faster, and reports results back.
The approach reduces latency and cost by offloading routine work from the most capable models.
The frontier model retains control over planning and judgment, while the specialist handles execution.
The system is designed to apply the right model to each part of a task, optimizing efficiency.