Quickstart

Get the Prismor runtime guard protecting your AI coding agent in about five minutes: install, run the setup wizard (which hooks your agent), enroll this machine, and verify it's working.

This walks through prismor, the runtime guard that watches your AI agent. If you're looking to scan a repo's dependencies for vulnerabilities instead, that's Patchbot, a separate open-source tool.


1. Check prerequisites

python3 --version   # need >= 3.8
pip3 --version
python3 -c "import yaml"   # PyYAML — required by the policy engine

If import yaml fails, install it first: pip3 install pyyaml.

2. Install

curl -sSL https://prismor.dev/install | sh

Or with pip directly:

pip install prismor

If pip refuses with externally-managed-environment (Debian/Ubuntu 23.04+, Homebrew Python), use pipx instead, then open a new shell so ~/.local/bin is on your PATH:

pipx install prismor
pipx ensurepath

3. Run the setup wizard

prismor setup

This is an interactive wizard: pick enforcement mode, choose a governance mode or which rules block, select which agents to hook, enable secret cloaking, pick which login (if any) judges uncertain events, choose install scope (this project vs. global), and optionally set an unlock password for the agent self-edit window.

Two modes, two philosophies:

  • Observe reports everything and blocks nothing, so setup leaves every rule on. This is the right first install.
  • Enforce first offers three named governance modes (dev-safe, trusted-workspace, regulated-airgap) or custom. Choose custom to pick rules one by one: it starts with nothing selected — including the safety floor, which is marked recommended rather than assumed. Nothing blocks until you choose it (press a to take the recommended set). Your selection is written to .prismor/policy.yaml one line per rule, so what blocks is readable in the file, and every block prints the exact prismor allow … command that makes a narrow exception. Prismor's own self-protection rules are the one thing not on the menu — they always block, because they are what stops an agent editing this selection.

For CI or scripted environments, skip the prompts:

prismor setup --non-interactive --mode observe
# or, enforcing with the recommended block set:
prismor setup --non-interactive --mode enforce --recommended
# or an explicit set:
prismor setup --non-interactive --mode enforce --enforce-rules destructive-command,secret-exfiltration

4. Enroll this device

If your team uses the Prismor dashboard, enroll this machine so it reports to your org's policy. Get a one-time enrollment token from Dashboard → Getting Started checklist → Enroll device (admins: Admin → Devices → Enroll a device), then:

prismor enroll <token>

The token is single-use and expires in 24 hours. Skip this step if you're only running Prismor locally with no org.

5. Check your agent is hooked

prismor setup already hooked the agents you picked, so there is no separate hook step. Confirm it:

prismor status
prismor doctor

To hook another agent later, run prismor install-hooks --agent <name> --scope project from inside the project. Agents: claude, cursor, windsurf, codex, copilot, gemini, antigravity, opencode, openclaw, hermes, grok, kiro, crush, openhands, qwen, continue, goose, or all.

Hooking Codex? Codex won't run a hook until you trust it: open codex interactively once in the workspace and accept the hook-trust prompt. Until then nothing is screened (prismor status and prismor doctor flag untrusted hooks).

Started in observe? Once you've reviewed a day or two of logs and trust the policy, switch to blocking:

prismor setup --non-interactive --mode enforce --recommended

Run install-hooks and setup in your own terminal, not through your agent. Once Prismor is hooked, its self-protection rule blocks an agent from changing Prismor's own configuration, and in observe mode it logs the attempt as a critical finding.

6. Verify it's working

prismor dashboard   # opens http://127.0.0.1:7070

Run a normal session with your agent, then check the dashboard — you should see tool calls appearing as events. If it's empty, you haven't run an agent session in a hooked workspace yet (see Troubleshooting).

If you enrolled the device in step 4, also confirm it shows up under Dashboard → My Policy with a recent check-in.


Next steps

Frequently asked questions

How long does it take to set up Prismor?

About five minutes. The quickstart walks through six steps: check prerequisites, install, run the setup wizard, enroll the device, hook your agent, and verify it is working.

What are the prerequisites for installing Prismor?

Python 3.7 or newer, pip3, and PyYAML, which the policy engine requires. If importing yaml fails, install it with pip3 install pyyaml before continuing.

What is the difference between prismor and prismor-cli?

prismor is the runtime guard that watches your AI coding agent and inspects tool calls before they execute. prismor-cli is a separate tool that scans repositories for vulnerabilities and secrets and generates an SBOM.

Do I need to enroll a device and install hooks separately?

Yes. Enrolling reports the device to your organization, but agent activity only appears after you run prismor install-hooks. They are two distinct steps in the quickstart.