Thuban Shield is local-first and runs the moment you install it. It gates every file write and every system command an AI coding agent tries to run — before it executes — across five layers: File Guard, Shadow Realm, Watchers, Execution Gate, and Audit & Recovery.
Thuban Shield sits transparently between the agent and the machine — visible, auditable, always on.
Real Thuban Shield output. Every action below was routed through the agent's write and execute calls and stopped by Thuban Shield before it reached disk.
Prompts and system messages are suggestions — an agent can ignore them, misread them, or be talked out of them. Thuban Shield is not a suggestion. It's a local process that intercepts file writes and shell commands at the OS boundary and decides, in real time, whether they're allowed to happen.
Everything runs on your machine. No file contents leave your workspace for Thuban Shield to do its job.
Every check happens before the write lands or the command runs — not in a nightly scan, not after the fact.
Works whether you're on Copilot, Cursor, Claude Code, Verdent, or whatever ships next month. Zero configuration.
Every layer runs locally. Every layer works whether you're on Copilot, Cursor, Claude Code, Verdent, or whatever ships next month.
Pre-emptive snapshots before any write lands. Protected paths (credentials, configs, git internals) are fenced off by default. Overwrite prevention stops an agent from silently clobbering files it didn't create.
A sandboxed execution environment that gives an agent a workspace to act in without direct, unmediated access to the host filesystem — so a hallucinated or malicious operation never touches production paths first.
Real-time filesystem monitoring that runs alongside the agent. Watchers flag credential-file access, unexpected directory traversal and out-of-scope reads the moment they happen — not in a nightly scan.
Every shell command an agent tries to run is validated before it executes. Known-dangerous patterns — mass deletion, privilege escalation, disabling monitoring — are blocked at the gate, not cleaned up after.
A hash-chained audit trail records what was attempted, permitted and blocked — tamper-evident by construction. If something does slip through, rollback is instant: restore from snapshot, not from memory.
Individual actions can look innocent. The chain reveals intent. Thuban Shield ships with five built-in chain detectors that watch sequences of actions across a session, not just one command at a time.
Zero-day philosophy: Thuban Shield doesn't try to predict every attack. It constrains what consequences are permitted. An agent may use an exploit nobody has seen before — but Thuban Shield can still see that it's touching credential files, leaving the workspace, spawning privileged processes, or trying to disable the very monitoring meant to catch it.
“I built Thuban Shield to fix my own problem: something local, real-time, that would sit between the agent and the machine and say no when it mattered. I open-sourced it because I figured other people were about to learn the same lesson the hard way.”
— Craig Lowther, Founder
Craig Lowther was running an AI coding agent on his own machine when it overwrote server.js in production — no warning, no confirmation, just gone. It wasn't malicious. It was just doing what it thought he asked, with far more authority than it should ever have had. He needed something that ran locally, watched in real time, and could sit between the agent and the machine to say no when it mattered. That's Thuban Shield.
Every protected session answers eight questions. Not vague reassurance — the actual events from your machine.
1. Which agents were active?
2. What did they attempt?
3. What was permitted?
4. What was blocked?
5. Which files were protected?
6. Were any secrets accessed?
7. Were snapshots created?
8. Anything to review?
Dashboard works even with telemetry completely disabled. Thuban does not ask you to trust a marketing claim — watch it work.
For every active agent session, Thuban Shield will maintain a live trust profile — not just a log of past events, but a running picture of how much this session should be trusted right now.
What this session is actually permitted to do right now.
Which paths this agent has touched or requested access to.
Every touch of a secret, key or credential file, permitted or not.
External domains and endpoints the agent has tried to reach.
Actions that don't match the pattern of the stated task.
Rules this session has tripped, resolved or still pending.
What a person has explicitly signed off on for this session.
A single number combining every signal above.
Enterprise agent security solutions are coming — behind procurement cycles, platform lock-in, and six-figure contracts. Thuban Shield is here now.
| Capability | Nothing | .gitignore | Container isolation | Thuban Shield |
|---|---|---|---|---|
| Works with any AI agent | — | — | depends on setup | ✓ |
| No configuration required | — | ✓ | ✗ | ✓ |
| Real-time, not after-the-fact | ✗ | ✗ | boundary only | ✓ |
| Behavioural chain detection | ✗ | ✗ | ✗ | ✓ |
| Instant rollback from snapshot | ✗ | ✗ | rebuild container | ✓ |
| Hash-chained audit trail | ✗ | ✗ | ✗ | ✓ |
| Credential-access detection | ✗ | hides files, doesn't watch | ✗ | ✓ |
.gitignore keeps secrets out of commits. It does nothing while an agent is running — it can still read, copy or transmit anything on disk. Container isolation raises the cost of an escape but adds setup friction and doesn't watch behaviour inside the boundary. Thuban Shield does both: contains the blast radius and watches what happens inside it, in real time, with zero configuration.
Real-time, local-first, zero-configuration protection — gating every write and command before it executes.
No install, no config files, no signup. Protect your first session in under 30 seconds — free.
npx thuban shield init
Activate Thuban Shield in the current workspace. Sets up File Guard, Shadow Realm, Execution Gate and Audit & Recovery immediately.
Get started free →
npx thuban shield status
See what Thuban Shield has protected so far in this workspace: permitted, blocked, snapshots created.
Get started free →
npx thuban shield rollback
Instantly restore the last protected snapshot if something slipped through.
Get started free →