Choose the shortest safe route into PALO-AI.

01 / Code-first

Validate the contracts and isolated runtime

Prerequisites: Node.js/npm, a local checkout and a supported shell.

1. Install exactly from the lockfile.
   npm ci Copy
2. Install the pinned local OPA tool.
   npm run opa:install Copy
3. Validate the agentic contracts and policies.
   npm run validate:agentic Copy

Expected result: schema, registry, policy and full-cycle reference checks complete without failures.

02 / n8n and visual builders

Sideload package 0.2 on a disposable canvas

Prerequisites: a disposable self-hosted n8n instance, Docker or an isolated host, and no production credentials.

1. Build and verify the package.
   cd packages/n8n-nodes-palo-ai && npm ci && npm run verify && npm pack Copy
2. Sideload the generated tarball.

   Install it only in your disposable self-hosted environment, then restart n8n.

3. Create the preview credential.

   For the local Docker demo use gateway http://host.docker.internal:8787 and the repository's explicitly demo-only token.

4. Import the safe workflow.
   examples/n8n-demo/PALO-AI-full-cycle-assurance-demo.json Copy

   Download the safe n8n demo JSON

5. Compare the nodes.

   PALO Governance is an advisory decision gate. PALO Governed Action represents the protected execution and verification path. Direct equivalent credentials remain a bypass and must be removed for real enforcement.

Expected result: visible Verified, Review Required, Denied and Execution Failed routes using synthetic data.

03 / Copilot Studio and MCP

Connect a deliberately narrow MCP toolset

Prerequisites: an isolated PALO remote MCP preview, HTTPS, a preview credential and permission to configure tools in Copilot Studio.

1. Start from one reversible action.

   Define its resource, allowed operation, accountable reviewer and expected effect.

2. Expose only the required PALO tools.

   Connect the dedicated Reader and verify its fixed six-tool catalog; do not publish the operational governance surface to the agent.

3. Use the Reader production identity boundary.

   Copilot Studio's wizard supports API key and OAuth 2.0. Use a short-lived OIDC access token with the exact Reader audience and two Reader scopes. Static bearer/API-key mode is limited to isolated local evaluation and is rejected by the production Reader.

4. Add the existing MCP server in Copilot Studio.

   Follow Microsoft's Streamable HTTP MCP onboarding flow. Use Reader for six read-only tools, or the separately authorized ten-tool Curator only for editorial workflows, then inspect every imported tool.

5. Remove direct equivalent actions.

   If the agent can still call the operational system directly, PALO is advisory rather than an enforcement boundary.

Expected result: the agent sees exactly six canonical read-only Reader tools. The code is a production candidate; the actual tenant remains pending until IdP, proxy and host qualification pass. Curator and operational actions retain their separate preview boundaries.