KonaSense secures the way employees and AI coding agents use generative AI at work. Security teams use it to see which AI tools are in use, approved or not, to apply policy to prompts and agent actions as they happen, and to keep audit evidence of what was allowed, changed or blocked.
The problem it targets is simple to state. Staff paste work into chat assistants, developers hand shells and repositories to coding agents, and the security team has little visibility or proof of compliance. KonaSense gives both human and agentic AI use a single control plane.
At A Glance
| Vendor | KonaSense |
|---|---|
| Category | AI security for browser AI use, AI coding agents and AI red teaming |
| Deployment | Managed Chrome and Edge extension; agent helper via MDM; SaaS, customer AWS or on-premises |
| Agentic AI | Pre-execution governance of shell, file, web and MCP actions by AI coding agents |
| Founders | Cristiano Lincoln, Rafael Silva, Felipe Zimmerle, Gustavo Monteiro |
| Market classification | AI Model Security; also AI Safety & Guardrails |
What KonaSense Does
KonaSense treats every AI interaction as a point where policy can be applied. Its work falls into three pillars: observability (usage analytics, anomaly detection, risk profiling), governance (policy control, shadow AI detection, data classification, compliance evidence) and security (protection against data exposure, prompt injection and unsafe AI behaviour). Each control follows the same four steps, See, Decide, Act, Prove: spot the interaction, decide against policy, act on it, then keep the record. The same loop covers people working in a browser, coding agents acting on a workstation, and models tested before they go live.
Products
Kona For Browser
Kona for Browser is for security and compliance teams worried about sensitive data leaving through ChatGPT, Gemini and similar tools. It classifies each prompt in real time as it leaves the browser, looking for sensitive data, prompt injection and risky intent, and can redact or anonymise content inline so secrets are replaced before anything reaches the model.
- Policy actions to block, redact, anonymise or coach the user, with nudges toward safer alternatives
- A live inventory of GenAI apps in use and an organisation risk score from 0 to 100
- Replayable evidence, policy logs and exportable records for audits
Kona For Agents
Developers increasingly let coding agents run commands and edit files on their own. Kona for Agents puts a decision point in front of those actions: shell commands, file operations, web fetches and MCP calls are checked before they execute and can be allowed, denied or redacted, or held for human approval. It supports Claude Code, GitHub Copilot, Cursor, Codex and Codex CLI, so one policy set governs several agents at once.
Detection runs pattern, semantic and classifier methods in parallel, and when those signals disagree an AI model arbitrates; customers can bring their own LLM keys for this. Each governed event is sealed in an ECDSA signed envelope, which gives tamper-resistant evidence of what the agent did.
KonaRed
KonaRed is adversarial testing for enterprises deploying models, agents and AI workflows. Teams point it at API endpoints, manual chat sessions, uploaded prompt and response pairs or agent workflows, then run curated attack packs or single scenarios. More than 31 scenarios cover prompt injection, data theft, agent abuse, identity impersonation, RAG poisoning, content safety and financial risk. Results are grouped by category and severity with pass or fail status, exploit examples and remediation guidance, and fixes can be retested.
How It Works
| Component | What it does |
|---|---|
| Browser extension | Kona for Browser ships as a managed Chrome and Edge extension rolled out through browser policies, with no proxy or endpoint agent. Prompts are inspected locally, and only policy decisions and redacted evidence leave the browser. |
| Agent hooks | Kona for Agents plugs into each agent’s official hooks rather than replacing the agent or sitting in the network path. Nothing stays resident on the developer’s machine; a small helper runs only while an action is being checked. It is a single signed executable distributed through MDM tools on macOS, Windows, Linux and ChromeOS. |
| Control plane | Policies are held centrally, never on the workstation, so a rule change applies to every supported agent without reinstalling. If the control plane cannot be reached, the action goes ahead and the agent shows a notice that policy was not checked. |
| Hosting | The control plane runs in KonaSense SaaS, in the customer’s AWS account or on-premises, with zero data retention by default and configurable per tenant. Customers can start in observe-only mode and switch on enforcement later. |
Integrations
| Area | Integrations |
|---|---|
| Browsers | Google Chrome, Microsoft Edge |
| Developer tools | VS Code |
| GenAI apps | ChatGPT, Claude, Gemini, Copilot, Perplexity and more than 50 GenAI tools |
| Device management | Intune, JumpCloud |
Company Background
| Leadership | Cristiano Lincoln (Founder and CEO), Rafael Silva (Founder and CPRO), Felipe Zimmerle (Founder and CTO), Gustavo Monteiro (Co-Founder and CRO) |
|---|---|
| Product and engineering | Rafael Silva leads product vision and AI research; Felipe Zimmerle leads architecture and engineering |
| Funding | Founder funded, with capital provided by Cristiano Lincoln |
Note: This overview is compiled from KonaSense’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.