Agentic governance partner pilot (Rewst / Neo)

External AI and automation tools are entering MSP stacks quickly. Rewst, Neo, ConnectWise zofiQ, and future MCP clients can propose remediation and open tickets. The durable risk is not the chat interface. The risk is unsigned or unattributable PowerShell landing on customer endpoints without an assessor-ready proof trail.

Trustholm's Path B answer is to remain the policy and evidence engine beside the incumbent RMM. External tools request privileged execution through Trustholm. Humans and machines share the same signing and approval gates. Every accepted intent produces attestation in the security audit plane.

This guide is the design-partner pilot checklist for Rewst or Neo sandboxes. Use it when you want one end-to-end attested run before you market agentic governance more broadly.

Why a partner pilot matters commercially

Regulated mid-market MSPs already keep Ninja, N-able, or ConnectWise for patch and remote. They will not rip and replace those platforms to adopt AI agents. They will ask a simpler procurement question: who is allowed to act, under which policy, with what proof?

A live partner pilot answers that question with artifacts rather than slides. Sales can attach an Assessor Package ZIP. Security can filter the AgenticExecution audit category. Leadership can see that Trustholm gates AI without becoming an L1 ticket bot competitor.

Outcome definition

Success for this pilot is narrow and measurable:

1. Agentic integration is enabled on a sandbox Govern (or Evidence) tenant.
2. A published, signed script is available to the agentic allowlist.
3. The partner submits POST /api/Governance/execution-intent with tenant code, integration key, and external actor id.
4. Approval completes when required.
5. A Windows agent runs the script.
6. Attestation events appear under category AgenticExecution.
7. An Assessor Package ZIP for the pilot window is retained for the internal business case.

If any step fails, stop and fix before expanding the partner scope. Volume without attestation is not a win for Path B.

Prerequisites

Prepare the sandbox before inviting the partner engineer:

• Govern tier is enough for a sandbox pilot. Evidence is useful when you need a longer Assessor Package window or higher row caps.
• At least one Windows agent must be online with a valid polling credential.
• Prefer a platform catalog script duplicated into the tenant library so signing and approval tags are realistic.
• Confirm whether RequireSignedScriptExecution is enabled. If it is, unsigned scripts will fail closed - that is correct behavior.
• Portal admin access is required for Settings → Integrations → Agentic so you can rotate the integration key once and store it in the partner secret store.

Optional: enable the Operations Intelligence beta SKU if you also want to prove suggestion approve → execution-intent with actor operations_intelligence. That bridge is useful for the rules-first intelligence story, but it is not required for the first Rewst or Neo run.

Step-by-step pilot

1. Create or select a non-production tenant that matches your sales demo story.
2. Install or confirm one agent on a lab machine that can receive PowerShell commands.
3. Publish a signed remediation sample (for example a harmless inventory or logging script). Avoid destructive scripts in the first pilot.
4. Open Settings → Integrations → Agentic, enable the integration, rotate the key, and copy it once.
5. From the Rewst or Neo sandbox, call execution-intent with headers X-Tenant-Code, X-Agentic-Integration-Key, and X-External-Actor-Id. Include script id, target agent ids, and a human-readable reason string.
6. If the intent returns pending approval, complete approval in the portal or through your configured approval workflow.
7. Poll intent status until completed or failed. Investigate agent connectivity if the intent stays queued.
8. In Security Centre or Audit, filter category AgenticExecution and confirm intent, approve, and attestation actions.
9. Generate an Assessor Package for the pilot date range. Confirm Govern caps (30 days / 10k rows) or Evidence caps when retention SKUs are present.
10. Attach the ZIP and a short write-up to the design-partner folder for Epic 5 case study follow-up.

Full HTTP examples live in the operator runbook under product docs (DOCS/06-Development/agentic-governance-partner-pilot-runbook.md). Product narrative for buyers: /platform/agentic-governance.

Troubleshooting table for partners

| Symptom | Likely cause | Next action | |---------|--------------|-------------| | 401 or 403 on intent | Wrong key, tenant code, or allowlist | Rotate key; confirm tenant header; check allowlist | | Signing rejection | Unsigned script under enforce policy | Sign and republish; or use a signed catalog duplicate | | Intent never runs | Agent offline or polling key mismatch | Fix agent runtime; verify queue | | No attestation row | Result not yet returned | Wait for agent completion; check dispatch linkage | | Export 403 window or rows | Govern caps exceeded | Narrow dates or upgrade Evidence retention SKUs |

Honest scope and non-goals

Shipped today: REST v1 execution-intent, attestation audit, portal settings, Assessor Package inclusion of the audit slice, rules-first Operations Intelligence bridge.

Roadmap: Native MCP transport over the same policy engine, outbound webhook callbacks when attestation completes, broader marketplace connectors.

Explicitly out of scope for this pilot: Replacing Neo or Robin as an L1 ticket resolution product, generative auto-execution of drafted scripts, Datadog-class root-cause chatbots, and RMM rip-and-replace positioning.

Commercial handoff after a green pilot

When the pilot is green, update the weekly metrics: agentic-attested executions, Assessor Package time-to-first-export, and whether the MSP is a Govern or Evidence candidate. Offer Evidence when questionnaire volume or retention needs exceed Govern caps. Keep Observe (monitoring/NOC) as an optional upsell - never the hero pitch in an agentic governance conversation.

Named customer quotes still require legal approval under Epic 5. Until then, use anonymized composites and the honest observation-in-progress SOC 2 language on the trust hub.