Codenotary launches AgentMon Start for small businesses

Codenotary has launched AgentMon Start for small and mid-sized businesses using autonomous AI agents in software development.

The software is designed to show how AI agents behave, what they access, and how much token spending they generate, without requiring a dedicated security team or cloud deployment.

Subscriptions start at USD $15 per user per month, with three development devices available free for six months at launch. The product is aimed at existing IT, development, and security teams rather than specialist AI operations staff.

AgentMon Start focuses on software development environments where AI agents can access files, write or change code, call tools and application programming interfaces, and make large numbers of decisions with limited human involvement. It tracks agent behaviour, use, and cost, while also offering policy controls and audit records.

Monitoring demand

The launch comes as businesses of all sizes bring AI agents into day-to-day operations. Codenotary argues that smaller organisations face the same basic oversight challenges as large companies, but often lack the staff and infrastructure to monitor autonomous systems closely.

"While corporations have the ability to set up AI governance teams, security operations centers and specialized observability infrastructure, non-enterprise organizations often cannot," said Moshe Bar, Chief Executive Officer and Co-Founder of Codenotary.

"That leaves a visibility gap, yet AI agents used by smaller organizations have access to many of the same sensitive systems, credentials, customer records and source code, and generate substantial cost," Bar said.

The product keeps AI activity within an organisation's own network and does not require a cloud account. That may appeal to businesses that want local oversight of development tools and data access.

The platform works across more than 16 AI agents and large language models, including Claude, Codex, Cursor, Gemini CLI, GitHub Copilot, Goose, Groq, KiloCode, Kiro, Ollama, and Roo Code.

Four areas

The product groups its monitoring into four areas: cost management; security and agent behaviour; incident investigation; and governance, including shadow AI detection.

For cost management, customers can track token consumption, model use, developer activity, session history, and spending across teams. The goal is to help organisations spot unusually expensive use, compare tools, and set budgets.

On the security side, AgentMon Start is designed to identify secrets appearing in prompts, monitor files accessed or modified by AI agents, detect possible prompt injection attempts, and record activity for later review.

For incident investigation, organisations can examine reasoning replay, execution traces, and searchable activity logs when an agent behaves unexpectedly. This gives teams a way to see what happened without piecing events together from separate application logs.

The governance tools are intended to show which AI applications employees are using, including unauthorised ones, while allowing policy controls to be applied across approved and unapproved tools.

Wider concern

Codenotary pointed to broader industry concern over how quickly autonomous systems can act without direct human review. It cited a July security incident involving an autonomous AI agent powered by OpenAI models that escaped a testing environment, reached the public internet, and compromised infrastructure at Hugging Face.

Bar said the episode underscored the need for direct monitoring inside companies using autonomous systems.

"The lesson from that incident is that autonomous software can take thousands of consequential actions before a human being could possibly understand what is happening. If you have agents operating inside your company, somebody needs to be watching them," Bar said.