Agent Permissions

No tools by default. You grant exactly what's needed.
Per-agent, per-directory access control that goes beyond what OpenClaw alone provides.

Pinchy agent permissions — allow-list with directory scoping

Allow-List, Not Deny-List

Most platforms start with "everything allowed" and ask you to block what's dangerous. That's backwards for AI agents that can read files and execute commands.

Pinchy starts with nothing allowed. A newly created agent is a blank slate — it can chat, but it can't access any tools, files, or external resources until you explicitly enable them.

❌ Deny-list approach

"Allow everything, block the dangerous stuff."

Problem: You have to anticipate every risk. Miss one tool, miss one directory, and the agent has access you didn't intend.

✅ Allow-list approach

"Block everything, enable what's needed."

Benefit: The worst case is an agent that can't do something. You add permissions as needed. No surprises.

Tool Categories

Pinchy organizes tools into two categories based on their risk profile:

🛡️ Safe Tools

Sandboxed access to admin-approved directories. The agent can list and read files — but only in directories you've selected. Every file access request is validated at runtime against the approved paths.

List approved directories Read approved files

⚡ Power Tools

Full capabilities that require more trust. Shell execution, web browsing, unrestricted file access. Enable these only for agents that genuinely need them, operated by users you trust.

Shell execution Web search Web fetch Full file system

Per-Agent Directory Scoping

This is the feature that makes Pinchy's permissions fundamentally different from OpenClaw's built-in tool configuration.

OpenClaw lets you say: "This agent can read files." Pinchy lets you say: "This agent can read files only in /data/hr/ and /data/policies/."

🤖 HR Onboarding Agent

✅ /data/hr/onboarding/ ✅ /data/policies/ 🚫 /data/finance/ 🚫 /data/engineering/ 🚫 Everything else

🤖 DevOps Helper

✅ /data/engineering/runbooks/ ✅ /data/engineering/configs/ 🚫 /data/hr/ 🚫 /data/finance/ 🚫 Everything else

The directory picker is visual — no path guessing, no config files. Select directories, save, done. The validation happens at runtime, not just at configuration time.

Permissions + Groups = Data Isolation

Agent permissions control what data an agent can access. Groups control who can talk to the agent. Together, they create real data isolation:

Agent Can access Available to
HR Assistant /data/hr/, /data/policies/ HR Group only
Finance Agent /data/finance/, /data/reports/ Finance Group only
Company Wiki /data/wiki/ Everyone

The finance team can't even see the HR agent. And even if they could, the HR agent can't access finance data. Two independent layers.

See Permissions in Action

30-minute demo. We'll set up an agent with scoped permissions and show you how data isolation works in practice.

Book a Demo