Kontext Raises $4M to Stop AI Agents From Going Rogue at Work
Kontext raises $4 million led by 42CAP, with backing from a16z CSX and HTGF, to expand its team and build runtime security for AI agents.
2026-09-24 · 5 min read
See every AI agent in use. Decide what they may do, before they do it.
Kontext is the runtime security platform for your AI workforce. Discover agents, map what they can access, block prohibited actions, detect risk in agent behavior, and manage spend to reduce costs. Keep evidence for every policy decision.
Kontext runs where your AI agents run. It maps access, detects risk and controls what each agent does and where data moves, so you can roll agents out safely across your business.
Kontext is the runtime security platform for your AI workforce. Discover agents, map what they can access, block prohibited actions, detect risk in agent behavior, and manage spend to reduce costs. Keep evidence for every policy decision.
Kontext protects your AI workforce, from coding assistants and workplace agents to custom-built agents, across endpoints and cloud environments. Bring agent discovery, access policies, risk monitoring, and spend management together across the agents your organization uses.
Agents act with your employees' access. Your tools can't see what they do.
These tools were built to check one event at one point. An agent's risk is in the chain of actions it takes with someone's access.
- Endpoint Security: Sees the process, not the action. A terminal is running. Not the command the agent chose, or why.
- Cloud Security: Sees the account, not the session. The posture of a service. Not what one agent session did with it.
- Identity & Access: Sees the sign‑in, not what follows. It knows who logged in. Not what the agent did next.
- Gateways: See one route, not the laptop. They cover the traffic that passes through them. Agents on a laptop mostly don't.
- Data Loss Prevention: Sees the pattern, not the agent. It matches content at an exit. Not which agent is moving it, or what it was asked to do.
Product
Agent Discovery & Access Posture: Know every agent on every endpoint.
Kontext finds the AI agents installed across your fleet, shows what each one can reach, and tells you where policy applies and where it doesn't.
- Finds coding and workplace agents, including Claude Code, Codex, Cursor, Claude Cowork, and Copilot CLI.
- Maps the tools, systems, and data each agent can reach.
- Shows which endpoints are governed, not governed, or not reporting.
Activity Monitoring & Risk Detection: See the risky actions, explained in plain language.
Kontext assesses each action an agent takes and flags the ones that matter. Every finding says what happened, who ran it, and why it is a risk.
- Findings read like a sentence, not a log line.
- Mapped to OWASP Agentic, OWASP LLM, MITRE ATT&CK, and MITRE ATLAS.
- Ranked by severity, person, and endpoint so you know where to look first.
Runtime Authorization: Turn a finding into a policy.
Start from presets for the systems agents touch, or let Kontext propose a policy from the risk it found. Watch it in observe mode, then enforce.
- Presets for GitHub, Jira, Confluence, Databricks, HubSpot, and the shell.
- Observe mode shows what a policy would have blocked before it blocks anything.
- Enforced before the action runs. Roll back at any time.
Spend Management & Optimization: See what agents cost, and spend less.
The same sensor that watches agent actions tracks agent spend. See it by person, model, and type of work, and let Kontext Compaction bring it down.
- Spend by person, agent, model, and type of work.
- Kontext Compaction cuts the tokens agents send to the model.
- Savings are reported per person and per day, not estimated.
What sets Kontext apart.
Detection runs on the device.
Kontext's detection models run locally on each endpoint. Actions are assessed in milliseconds.
Risk becomes policy.
Other tools hand you a queue of findings. Kontext turns a finding into a policy you can test in observe mode and then enforce.
Security that lowers the bill.
Kontext tracks agent spend and cuts it with compaction. The rollout that reduces risk also reduces cost.
Discover what others have to say about Kontext
Kontext has already flagged risky agent activity during our evaluation. Being able to search commands and connections across users helps us investigate what agents are doing and identify where our security team needs to take a closer look.
After thoroughly evaluating Kontext, I can attest to its lightweight design, its impressive engineering around local detection models, and the great team behind it, who keep delivering steady improvements and updates. It has become my go-to recommendation whenever the conversation turns to shadow AI and endpoint AI risk.
Rolls out through the tools you already run.
Kontext ships as one signed package for macOS, Windows and Linux. Push it with your MDM and endpoints start reporting. No kernel extension, no code changes, and nothing changes for the people using the agents. Deploys with Jamf, Microsoft Intune, Iru, JumpCloud, Addigy, and any MDM that can push a .pkg or .msi.
- Deploy: Roll out to every endpoint through your MDM.
- Observe: Agents, their access, and their activity appear in the dashboard.
- Validate: Review what a policy would have blocked before it blocks anything.
- Enforce: Switch policies on. Risky actions are stopped before they run.
Blogs
Merlin: Context-aware Local Detection of Unsafe Agent Tool Calls
Merlin evaluates agent tool calls locally using the user request, tool history, and tool schemas to detect unsafe actions before execution.
2026-09-21 · 8 min read
Guides on AI agent security
Pricing
Free: $0; 5,000 transactions a month. Team: $500/month; 100,000 transactions, then $1 per 1,000. Enterprise: Custom; Unlimited transactions.
Enterprise pricing is custom. Contact sales for a rollout tailored to your organization.
Sign up, give a coding agent the setup runbook, book a demo, or talk to sales.