Give every agent an identity and enforce your rules on every request it makes.
Works with the agents your teams already use, without changing them.
Product
Endeem gives each agent an identity, applies your rules to its requests and keeps a record of all of it.
Your rules decide first. Anything they don’t cover is checked against what the user asked for.
Jane asked Claude Code
“Fix the failing login test and open a PR”
Endeem checks on the machine which agent is running, who signed it and who it works for, including shadow AI.
Identity verified
Credentials are injected just-in-time, for one request, and never handed to the agent.
POST …/pullsno credentialGitHub tokenthis request onlynothingEndeem suggests rules from what your agents do. You approve them instead of writing them.
Suggested rule
Block DROP and DELETE on prod-db for all agents
Endeem reviews what your rules allowed once it has run. If an action was out of line, it stops the session, blocks the agent and revokes its access.
Out-of-line action
See each LLM call, tool call and subagent in one view, tied to the agent and the employee who started it.
Why Endeem
Agents act with your people’s access, and most of what they do never passes a gateway.
Claude Code runs gh, aws and kubectl with Jane’s own login. Endeem checks the running process and knows the request came from Claude Code, working for Jane.
Commands, scripts and direct API calls go around it. Endeem sees each request as it leaves the machine.
To a hook, python cleanup.py is one harmless command. Endeem sees the two deletes it makes and decides each one.
How it works
Endeem runs on the same machine as the agent. Shadow AI, or an agent that skips its hooks, still goes through it.
Claude Code, Codex, Claude Cowork and Microsoft Copilot on your people’s machines.
Your own agents on VMs, Kubernetes and Bedrock AgentCore.
A hosted Enforcement Point for agents a vendor runs for you.
Solutions
Security keeps control, so everyone else can say yes to agents.
FAQ
No. Endeem works with coding agents like Claude Code and Codex, and assistants like Claude Cowork and Microsoft Copilot, with no changes to the agent or to how your teams work.
A hook sees the command the agent chose to run, like python cleanup.py. Endeem sees each request that command sends, so a script the agent writes and runs a second later is still checked, request by request. Hooks also only work in agents that support them and assume the agent harness is trusted. Endeem doesn’t need the agent’s cooperation.
A gateway only sees traffic routed through it. An agent that changes its endpoint, uses a personal key or calls an API directly goes around it. Endeem runs on the same machine as the agent, so those requests still go through Endeem.
Endeem identifies the agent from its running process on the machine, including who signed it and who started it. It doesn’t rely on what the agent says about itself or on a directory entry.
Your rules run on the device and keep enforcing. Anything your rules don’t decide is blocked until the intent check can answer again.
No. A block only applies to the agent’s requests. The employee keeps their own access and can keep working.
We’ll show you how Endeem identifies your agents and decides each request they make.
We’ll reply by email to find a time.