Skip to main content

PALO Guide | Deterministic local workbench

Understand PALO. Find the next accountable step.

Describe the governance situation, see which signals changed the route, and leave with concrete artifacts and a bounded product integration pattern.

Find my route Plan an MCP integration

Browser-local and source-grounded. This guide does not send answers, make legal decisions, certify compliance or authorize deployment. Answers remain in this tab unless you explicitly choose device storage.

Guided workbench

Turn six signals into a traceable route.

Change an answer and run the route again. The rules are local and deterministic, so the same inputs produce the same recommendation.

Route output

Recommendation

Local inference

Ready for your inputs.

01-06

Start with the situation, not a module name.

The route will show a starting phase, two to four PALO steps, expected evidence and an MCP integration class.

How PALO reasons

Six phases, one evidence thread.

The governance loop is canonical. The recommendation selects a proportionate entry point; it does not skip accountable review or turn framework guidance into authority.

  1. 01Frame

    Define purpose, scope, owners and affected people.

    Artifact: use-case brief
  2. 02Classify

    Establish the risk route and level of scrutiny.

    Artifact: risk route
  3. 03Assess

    Evaluate impacts, authority, autonomy and oversight.

    Artifact: impact record
  4. 04Control

    Assign owned, testable safeguards and gates.

    Artifact: control plan
  5. 05Measure

    Set indicators, thresholds, owners and cadence.

    Artifact: KPI/KRI register
  6. 06Prove & Review

    Assemble evidence and decide what must reopen.

    Artifact: evidence bundle
Evidence is not authority by implication.

Canonical definitions structure the work. Source-backed context still needs applicability review. Local previews are not operational evidence. Human-review-required signals cannot change governance state on their own.

Integrate PALO into your product

Guide first. Protect actions separately.

The dedicated six-tool Knowledge Reader explains, plans, searches and retrieves only digest-bound canonical PALO records. Its stateless production candidate requires live deployment qualification. The restricted ten-tool Curator adds immutable draft and review operations under a separate boundary. Neither profile authorizes deployment or executes protected actions; add the PALO-AI governed path only when a product can cause effects.

Product edge Client, product or workflow Authenticated by the product
PALO guide tools Explain, infer, plan Read-oriented governance guidance
Released sources Semantic, gate and control registries Versioned definitions and boundaries
When actions need protection
Action claimPALO-AI governed pathProtected executor and verified effect
The developer-preview runtime is not a production authorization boundary. Use isolated, non-consequential tools until readiness gaps are closed.
palo_explain_framework

Explain phases, artifacts, boundaries and released framework concepts.

palo_infer_governance_route

Return a deterministic, reasoned route from declared context signals.

palo_plan_product_integration

Recommend a bounded integration class, architecture and readiness cautions.

Read the PALO Guide Agent and MCP integration manual for the prompt contract, tool inputs, response boundaries and product-side implementation checklist.

Local development

MCP over stdio

{
  "mcpServers": {
    "palo": {
      "command": "node",
      "args": ["/path/to/PALO/packages/palo-mcp-server/index.js"]
    }
  }
}

Use an absolute path and isolated development data. The reference package remains a Developer Preview.

Remote Reader

Authenticated Streamable HTTP

{
  "type": "streamable-http",
  "url": "https://your-palo-host.example/mcp-guide",
  "headers": {
    "Authorization": "Bearer ${PALO_MCP_TOKEN}"
  }
}

The Reader exposes exactly six knowledge tools. Production mode accepts only short-lived OIDC access tokens for the exact resource audience and requires TLS, the fixed two-scope role, an explicit host allowlist and the canonical-only release.

Configuration examples are ready to copy.

Eleven host paths, one qualification rule

Paired Reader and Curator configurations are available for Copilot Studio, Claude, ChatGPT/Codex, GitHub Copilot, VS Code, Cursor, Gemini, JetBrains, AWS AgentCore, n8n and Dify. Repository validation covers protocol and configuration; every tenant remains pending until a live authenticated test is recorded.

Open the Knowledge Copilot integration matrix
Production boundary check

The Reader code is a production candidate for informational Q&A; IdP, proxy, logging, monitoring, rollback and host checks remain deployment responsibilities. Curator and protected-action PALO-AI retain their separate non-production boundaries.

Review the Reader production profile

Continue with a working surface

Move from orientation to an artifact.

Build evidenceEvidence PackProduce a local, reviewable evidence dossier. Bound delegated actionPALO-AM simulatorTest action space, autonomy and reversibility. Inspect the control planeGovernance HubExplore the developer-preview workflow and evidence model. Implement the guide agentGuide and MCP manualUse the deterministic prompt, tools and integration classes. Check the boundaryProduction readinessSee remaining controls and current evidence status.