We secure your agents
SixWays to Sunday.
Security and governance for AI coding agents, on the laptop and at the network edge.
One policy wherever your agents run: a laptop, a container, a microVM, NVIDIA OpenShell, or a cloud sandbox. The Endpoint is free. Add fleet policy, approvals, an AI firewall, and evidence when your team needs them. SixWays is in early access.
- ✓ Open source
- ✓ No account, no server required
- ✓ Telemetry stays on your machine
Why existing security tools miss AI agents.
Agents act with your rights. A coding agent can read every file and secret, run any command, call any host, and push to git. Each existing tool sees only part of that.
Endpoint security
Sees processes on the host, but not the instructions the agent was given.
Prompt filters
Check the conversation, but not what the agent then does on the machine.
Built-in guardrails
Set up separately in each agent, and an injected instruction can get the agent to bypass them.
SixWays applies one policy to what the agent does on the machine and what it sends to the model, and enforces it outside the agent process.
Six ways to secure an agent.
Three protect your machines from what agents do. Three counter attacks that manipulate agents.
-
Monitor
eBPF records file, process, and network activity from a sidecar the agent cannot see or tamper with.
-
Contain
Hardened containers, process sandboxes, or microVMs. A compromised agent can reach its workspace, not the rest of your machine.
-
Harden
Minimal images, read-only rootfs tiers, no sudo, and policy translated into each agent’s own settings.
-
Intercept
The AI Firewall scans prompts and responses in-flight, before an attack reaches the model or data leaves it.
-
Govern
One policy from org to endpoint; lower levels can only add restrictions. MCP tool calls are checked before they run, and risky changes wait for a person to approve them.
-
Sever
On a violation, SixWays blocks the connection, stops the process, and keeps a tamper-evident record.
One platform, from laptop to fleet.
One policy and one audit trail across 26 agents, three operating systems, three isolation levels, the AI Firewall, MCP tools, and git. Each component works on its own, so adopt only what you need.
How the pieces fitSixWays Endpoint
Run any coding agent in a sandbox you can see inside. Open source.
SixWays Server
Fleet policy, telemetry, and secrets in one console.
SixWays Firewall
Scans every prompt and response between agent and model.
SixWays Sandbox
Hardened Wolfi images, public on ghcr.
SixWays Pulse
Signed profiles of expected file, process, network, and tool activity.
SixWays AppSec
Scans the agent’s code before the commit leaves the sandbox.
For NVIDIA OpenShell
Keep OpenShell. Add one policy, approvals, inspection, and proof.
SixWays Arsenal
Adversarial test cases for the models you depend on.
IDE extension
Free. Attributes every operation to the agent that caused it.
Deeper isolation, when you need it
MicroVM isolation, process sandbox, git governance, the Trusted MCP Catalog, zero-trust mesh, and Cloud Sandbox use the same endpoint and policy.
What's free and what's paid.
The free build has no artificial limits. Some components are closed-source binaries or commercial.
Open source: the endpoint
Daemon, CLI, TUI, desktop app, container sandbox, eBPF monitoring, policy engine, session recording, signed updater.
Free binaries: eBPF enforcement and the process sandbox
Both are closed source. The eBPF enforcement sidecar ships as a prebuilt image with no license key, and the OS-native process sandbox is free for local use.
Commercial: Pro and the fleet layer
Pro gives one developer Pulse, the Trusted MCP Catalog, microVM isolation, and local session undo. Team and Enterprise include Pro and add the fleet: Server, git governance, OpenShell governance and approvals, and add-ons such as the Firewall, AppSec, and mesh.
Get early access.
Tell us about your agents and your team, and we will set you up.