# AI use cases for Aldus Palace (heymi/aldus-palace)

Repo: [heymi/aldus-palace](https://github.com/heymi/aldus-palace) | Use cases: 4 | Page: https://dedupedai.com/use-cases/heymi-aldus-palace/

A SQLite-backed context layer that turns free text into typed commitments, decisions and memories, each with its evidence, exposed over MCP, HTTP, a library and a typed client.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Give Claude and Cursor one shared memory file: https://dedupedai.com/use-cases/heymi-aldus-palace/shared-memory-for-claude-and-cursor/
- Sort user sentences into tasks and memories in code: https://dedupedai.com/use-cases/heymi-aldus-palace/sort-free-text-into-typed-records/
- Run a self-hosted capture API with Docker: https://dedupedai.com/use-cases/heymi-aldus-palace/self-host-capture-api-with-docker/
- Script a check on actions the gate is holding: https://dedupedai.com/use-cases/heymi-aldus-palace/script-review-of-gated-agent-actions/

## How to give Claude and Cursor one shared memory file with Aldus Palace

Repo: [heymi/aldus-palace](https://github.com/heymi/aldus-palace) | Difficulty: Intermediate | Time: 15 minutes | Basis: From the README

Install the MCP server so your assistant stores commitments and remembered rules in one SQLite file.

**Who it is for:** An individual who uses Claude or Cursor daily and finds that each client forgets what the others were told.

**The problem:** Your preferences and to-dos stay locked inside one assistant app, so other clients cannot see them.

**The outcome:** One MCP-backed store that your clients read and write, where each memory shows the sentence it came from.

### Steps

1. Run `npm install -g @aldus-palace/mcp`
2. Run `claude mcp add aldus-palace -- node "$(npm root -g)/@aldus-palace/mcp/dist/index.js"`
3. In Claude, send a sentence such as: I prefer simple tools
4. Confirm the reply says the memory is active and gives its evidence
5. Open Cursor with the same MCP server configured and ask what it has captured

### Prompt to paste into your agent

```text
Using the repo heymi/aldus-palace, install the MCP server with `npm install -g @aldus-palace/mcp` and register it with Claude Code using the command in the README. Send the sentence 'Ship the <target> next week' and expect a reply beginning 'Captured' with one commitment. Verify by checking that the reply shows a resolved date window.
```

### Caveat

The README's offline demo uses the deterministic provider, so connect a model if you need general understanding beyond the sentences it knows.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heymi-aldus-palace/shared-memory-for-claude-and-cursor/

## How to sort user sentences into tasks and memories in code with Aldus Palace

Repo: [heymi/aldus-palace](https://github.com/heymi/aldus-palace) | Difficulty: Advanced | Time: 45 minutes | Basis: From the README

Call the core library so a product turns each user sentence into a commitment, decision or memory.

**Who it is for:** A developer building an AI product where users type plain sentences and the product needs structured records.

**The problem:** Writing classification rules for tasks, decisions and preferences by hand is slow and brittle.

**The outcome:** Your app stores each user sentence as written and gets back a summary card with typed objects and resolved dates.

### Steps

1. Run `npm install @aldus-palace/core` in your project (inferred, check the README)
2. Copy the library example from the README into a file, for example capture.ts
3. Run it with `npx tsx capture.ts` or your existing TypeScript runner (inferred, check the README)
4. Check that the printed card reads 'Captured · 1 commitment' for the onboarding sentence
5. Replace the sample sentence with your own user input and repeat

### Prompt to paste into your agent

```text
In a project that uses the repo heymi/aldus-palace, install @aldus-palace/core and run the library example with a sentence of your choice. Expected output: a formatted action card naming the record type and any resolved date window. Verify that the database file ./aldus.db was created.
```

### Caveat

The README example runs on the offline deterministic provider, so output differs from a live model until you connect one.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heymi-aldus-palace/sort-free-text-into-typed-records/

## How to run a self-hosted capture API with Docker with Aldus Palace

Repo: [heymi/aldus-palace](https://github.com/heymi/aldus-palace) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Start the HTTP service in Docker so your own frontend or scripts can post sentences to one store.

**Who it is for:** A small team or developer who wants a single capture endpoint backed by one SQLite file in a volume.

**The problem:** Each new client needs its own capture logic and storage, which duplicates work.

**The outcome:** A running HTTP service that accepts a sentence and returns a captured record.

### Steps

1. Run `cp .env.example .env && docker compose up --build`
2. Change the bearer token in .env from the example value before exposing the port
3. Run `curl -X POST localhost:8787/v1/inputs -H 'Authorization: Bearer dev-local-token' -H 'Content-Type: application/json' -d '{"content":"Ship the onboarding page next week","mode":"sync"}'` (inferred, check the README)
4. Check that the response includes the captured commitment
5. Point your frontend at the same endpoint

### Prompt to paste into your agent

```text
In the repo heymi/aldus-palace, copy .env.example to .env, replace the bearer token, and start the service with docker compose up --build. Send a POST to /v1/inputs with a sentence of your choice. Expected output: a JSON response describing the captured record. Verify that the response reports a status and that a second request with a repeated sentence points to the first record.
```

### Caveat

The README says to run one instance per person, so a shared team deployment needs separate instances rather than one shared token.

Risk, credentials: The README example uses the placeholder token dev-local-token, which must be replaced before the port is reachable from other machines.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heymi-aldus-palace/self-host-capture-api-with-docker/

## How to script a check on actions the gate is holding with Aldus Palace

Repo: [heymi/aldus-palace](https://github.com/heymi/aldus-palace) | Difficulty: Advanced | Time: 45 minutes | Basis: From the README

Use the typed client to capture a sentence and list proposed agent actions from a script.

**Who it is for:** A developer or ops engineer who lets an agent propose actions and needs a script to review what is waiting for approval.

**The problem:** Proposed agent actions wait in a gate, and checking them by hand through a UI is slow.

**The outcome:** A script that captures input and prints the actions currently in the proposed state.

### Steps

1. Run `npm install @aldus-palace/client` in your project (inferred, check the README)
2. Start the HTTP service from the self-hosted Docker step
3. Copy the typed client example from the README into a script file
4. Run the script and read the printed card summary
5. Read the list of proposed actions the script prints and decide which to approve in your own tooling

### Prompt to paste into your agent

```text
In a project using heymi/aldus-palace, install @aldus-palace/client and create a client with baseUrl http://localhost:8787 and the token from your .env. Capture the sentence '<target task>' and then request the proposed actions. Expected output: a summary line and a list of proposed actions. Verify that the summary line matches the README format 'Captured · <n> commitment'.
```

### Caveat

The README says the runtime records intent and plans, and you decide what acts on the world, so the script only reads gated actions and does not approve them.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heymi-aldus-palace/script-review-of-gated-agent-actions/