Guardrails for AI coding agents
AI agents ship real work and real mistakes. NebGuard sits between your agent and the actions it takes, checks every one against a signed rule library, and records the decision, so Claude Code and Codex move fast without taking your production down with them.
Four Answers to Every Action
On each action your agent is about to take, NebGuard answers one of four ways. No cloud round-trip, no added latency worth noticing.
Allow
Routine actions proceed untouched. NebGuard stays out of the way of normal work, so the agent keeps its speed.
Guide
Risky actions run, but the agent receives a note about what to watch, steering it toward the safe path without stopping it.
Block
Dangerous actions stop before they execute: force-pushes to protected branches, destructive cloud calls, leaked credentials.
Bypass
The agent can proceed on a flagged exception, but the bypass and its stated reason are written to the audit trail, so even an override stays on the record.
Installed in a Minute
Signed, verified binaries for macOS, Linux, and Windows. The one-line installer even detects your AI agents and wires them for you.
$ curl -fsSL https://get.nebcore.ai/install | bash -s -- nebguard
One line on macOS, Linux, and Windows shells (WSL or Git Bash). Signed end to end, and it detects Claude Code and Codex and wires them for you.
Prefer to read it first? The installer is a plain shell script: curl -fsSL https://get.nebcore.ai/install | less
Then wire your agent
Claude Code
nebguard setup claude-codeCodex CLI
nebguard setup codex-cliThe curl installer runs these for detected agents automatically; use them directly any time. Setup starts your free 7-day trial of the full rule library; the free baseline keeps working forever after.
Signed Rule Packs, Six Domains
Rules ship as signed, encrypted packs spanning security, development, devops, ITSM, AI governance, and business content. Most packs are free.
Coverage that matches your stack
Add what fits, skip what does not, and update from the CLI as your stack changes.
Signed, encrypted, verified
Every pack is signature-verified before it loads, and rules ship as data, so coverage evolves without waiting for a new binary release.
Your rules alongside ours
Write custom rules with schema-validated YAML, layer organization-level overrides, and manage all of it from the local dashboard.
See What It Caught
One command, nebguard serve, opens a local dashboard reading your own audit trail. Everything below is real: lifetime counters and decisions from one of our own development machines.
1,088
BLOCKED
3,456
GUIDED
$18,718
EST. SAVED
51
SESSIONS
rm -rf "$DIR"unexpanded variable · criticalgit commit -m "feat!: …"unintended major releasework declared doneno test-run evidence · stop gatenpm run testroutineBLOCKOne unset variable from disaster
rm -rf "$DIR"If the variable is empty, this expands to the filesystem root. Stopped before it ran; the agent re-issued it with guards.
$18,718 saved
1,088 incidents that never happened, estimated from what each block avoided. Lifetime, one machine.
An Audit Trail Your Auditors Can Use
Every decision writes a tamper-evident audit record: the rule, the decision, the severity, and the tool it applied to. Evidence for SOC 2, HIPAA, and NIST controls falls out of normal operation instead of a quarterly scramble.
And it is local-first: decisions never leave your machine, and telemetry is opt-in: it never carries code, prompts, or secrets, and file paths and command arguments are redacted.
Built so the guarantees hold when nobody is watching.
Frequently Asked Questions
Is NebGuard free?
Yes. Installing is free and the baseline guardrails keep working forever. Every install starts a 7-day trial of the full rule library; after it ends, subscription rules pause until you subscribe, and your work is never blocked by their absence.
Which AI coding agents does it support?
Claude Code and OpenAI Codex CLI today, with more hosts on the roadmap. One setup command wires the hooks: nebguard setup claude-code or nebguard setup codex-cli.
Does my code leave my machine?
No. Decisions are evaluated locally and stay on your machine. Telemetry is opt-in; it never carries your code, prompts, or secrets, and file paths and command arguments are redacted before anything leaves your machine.
Do I need the NebCore AI Platform to use it?
No. NebGuard works standalone. If you do use the platform, signing in with NebCLI activates your subscription guardrails automatically, and the same rules govern the platform's own autonomous agents.
Put a seatbelt on your AI agents
Install in a minute, watch the first guardrail fire, and keep the free baseline forever.