DIP — Decision Governance

Execute governed business decisions via AI agents. Agents draft rule changes; humans always approve.

Serveur MCP hébergé

npx add-mcp 'https://dip.minhdangco.com/mcp'

S’installe dans Claude Code, Codex, Cursor et plus

Documentation

One run, step by step · Refund approval · v7synthetic data

  1. Receive refund request · all required fields present Valid
  2. Check return policy · within period · eligible category Pass
  3. Look up order history · source did not answer → missing data Unknown
  4. Compute automatic refund · missing input from step 03 Unknown
  5. Conclusion · by version v7 · trace saved Send to reviewer

The order history source did not answer, so the refund amount could not be computed. Missing data is recorded as missing, never treated as a pass, so the case goes to a reviewer.

trace · 5 steps · saved, replayable

Three familiar situations

The same decisions are made every day, yet the rules are scattered across source code, spreadsheets and people’s memory.

  • Changing a threshold means waiting for IT

    Approval conditions live in code or spreadsheets, so every change becomes a development request.
  • Nobody can explain a rejection

    When a customer asks why, the team has to retrace every step by hand.
  • Nobody knows who changed what

    Rules change many times, with no record of who edited them, who approved them, or when.

Four steps, all the way to “why”

  1. Design

    Build the process as a diagram and condition sets — as a table or imported from Excel. Each step does one thing.
  2. Test

    Run it on sample cases and test suites, and see which rules are still untested before anything goes live.
  3. Approve & release

    The author and the approver are two different people. The approver sees exactly what changed.
  4. Run & explain

    Every run keeps a full trace. See why it reached that result, in plain language.

Let an AI assistant run processes or draft changes — a person still approves.

Connect Claude, ChatGPT or your own agent to DIP over MCP. Ask about a specific case and the assistant calls the deployed version instead of guessing. To change a rule, it creates a draft, tests it on sample cases and leaves it for another person to approve.

  • Runs deployed processes, every run traced
  • Drafts and tests condition sets and processes
  • Cannot approve, release or deploy; whoever drafts cannot approve
  • Uses the user’s own permissions, personal data masked

AI assistant · connected to DIPsynthetic data

  1. You Add a condition: refund requests above 5M VND go to a reviewer. Test it for me.
  2. create_rule_set_draft Refund policy · from v7
  3. test_rule_set_version v8 · test cases
  4. Draft v812/12 test cases pass · awaiting another person’s approval
  5. Assistant I created draft v8 with the new condition and ran 12 test cases; all of them pass. The draft needs another person to approve it in DIP before it takes effect.

Wherever decisions repeat every day

Your trial comes with a sample process that matches the need you selected.

  • Online shops

    Approve returns and refunds: is the order eligible, how much to refund, which cases go to a reviewer.
  • Customer support teams

    Classify and route requests, cap compensation according to your own policy.
  • Application review units

    Is the application complete, what is missing, who reviews it next.

Sales teams approving discounts: discount approval rules you can test and explain

Who did what, when, and under which version — all on record.

  • Every run traced

    Inputs, each step that ran and the result are stored so you can review and replay them.
  • Two-person approval

    The system never lets one person both author and approve the same version.
  • Immutable versions

    To change a rule, create a new version. The old one stays intact for comparison.
  • Personal data masked

    Fields marked as personal data are masked in logs and displayed results.

The trial runs on a shared server and is for synthetic data only — do not upload real case files or customer data. AI, when enabled, is only an optional evaluation step; a person still makes the decision. AI assistants connected over MCP can only run processes that have been approved and deployed and only create drafts, with exactly the signed-in user’s permissions, and cannot approve, release or deploy.

Integrate DIP into your systems

Call decisions over an API, get results with explanations, and handle missing data correctly. Documentation in Vietnamese, SDKs for.NET and TypeScript.

  • REST API — call it from your systems, with Idempotency-Key and webhooks.
  • SDK — client libraries for.NET and TypeScript.
  • AI assistants over MCP — Claude, ChatGPT or your own agent run deployed processes and draft changes with the user’s permissions; every action is on record.
# example, synthetic data
POST /v1/decisions/{key}/execute
{
  "input": { "transaction": { "amount": 60000000 } }
}

200 → conclusion, version, trace and explanation

# or let an AI assistant call it over MCP
claude mcp add --transport http dip https://dip.minhdangco.com/mcp
execute_decision → same result, personal data masked

Try it on a sample process

Fill in a short form. We review every request by hand, create a private workspace with a sample process, and send your account through a separate channel. The trial is for synthetic data only.

  1. You Send the trial form
  2. We Review the request by hand
  3. You get A private workspace with a sample process