Canyon Code, a Sunnyvale, California startup that builds infrastructure for companies running several AI agents together, emerged from stealth on 26 May 2026 with a $5m pre-seed round. Cota Capital led the round, with participation from Newbuild Ventures and Blackhorn Ventures. SiliconANGLE named the firms Newbuild Ventures and Blackhorn Ventures, while The SaaS News used the name Newbuild Venture Capital for the former.
No valuation was disclosed, and the reports did not name any earlier funding.
“Enterprises are crossing the dependability thresholds with agentic systems and are beginning to deploy more and more multi-agentic apps at scale,”Ravikiran Gopalan, co-founder and CEO, Canyon Code
The problem Canyon Code targets
SiliconANGLE described the issue this way: companies are starting to move from experimental pilots to large multi-agent applications, but the infrastructure under them was built mainly for the layer that serves AI models, not for the workflow of agents on top. As a result, teams lack visibility into what their agents do, the actions they take and the decisions they reach.
Canyon Code co-founder and chief executive Ravikiran Gopalan said that lack of clarity is behind the spiralling cost and inconsistent performance of agent workloads. He argued enterprises need a workflow intelligence layer that can show and manage the collective behaviour of their agent applications.
What the product does
Canyon Code's product sits above the model-serving layer. According to SiliconANGLE, its central component is a dependence graph that tracks the dependencies and interactions between agents and follows each agent's progress in real time. That information goes back to the model-serving layer so that model calls can be scheduled more efficiently.
The report gave an example: if one agent's output is needed by three others, the system can prioritise that agent's call to cut delay overall. The software also manages contextual memory, so each agent gets the information it needs without larger prompts that raise costs.
The company's website describes three functions: observe, which shows what each workflow costs and where it stalls; optimise, which routes and schedules model use; and govern, which sets per-workflow and per-user-persona controls.
Gopalan said a company could, for instance, set customer-facing support agents to favour low delay while back-office data-analysis agents favour accuracy and cost, even when both run on the same model.
Founders
The company was founded by Gopalan and Aditya Akella, who The SaaS News said is a professor. SiliconANGLE described Gopalan as a three-time founder who has already scaled one agentic AI startup, and Akella as a researcher who has written papers on machine learning and operating systems. The sources disagreed on the founding year, so it is not given here.
Gopalan said enterprises are crossing the dependability thresholds with agentic systems and are starting to deploy more multi-agent applications at scale, but have no easy way to set behaviour policies on a per-app and per-persona basis.
What the money is for
Canyon Code said it will use the money to build out the workflow intelligence layer and to hire across research and development. SiliconANGLE said the funds will help accelerate the startup's research.
The reports did not give a customer count, revenue or headcount.
Company website: canyoncode.ai