# AI use cases for AMA (Sherlockwz/AMA)

Repo: [Sherlockwz/AMA](https://github.com/Sherlockwz/AMA) | Use cases: 4 | Page: https://dedupedai.com/use-cases/sherlockwz-ama/

AMA is a Python and TypeScript research codebase and agent memory skill that gives agents long-term memory through raw text, fact, and episode stores.

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

- Add AMA long-term memory policy to an agent harness: https://dedupedai.com/use-cases/sherlockwz-ama/harness-agent-memory-skill/
- Give an OpenClaw agent automatic recall and capture: https://dedupedai.com/use-cases/sherlockwz-ama/openclaw-memory-plugin/
- Start the AMA memory sidecar and confirm it responds: https://dedupedai.com/use-cases/sherlockwz-ama/local-memory-sidecar-smoke-test/
- Reproduce AMA's LoCoMo and LongMemEval results: https://dedupedai.com/use-cases/sherlockwz-ama/reproduce-am-paper-benchmarks/

## How to add AMA long-term memory policy to an agent harness with AMA

Repo: [Sherlockwz/AMA](https://github.com/Sherlockwz/AMA) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Load the portable SKILL.md policy and bind AMA memory operations to your agent's tools.

**Who it is for:** An agent developer whose harness forgets user details between sessions and needs a documented policy for when to recall, capture, and forget.

**The problem:** Agents lose prior context and have no consistent rule for when to retrieve memory, store turns, or honor deletion requests.

**The outcome:** Your harness loads a written memory operating policy and can call AMA memory operations through its own tool interface.

### Steps

1. `git clone https://github.com/Sherlockwz/AMA.git`
2. `cd AMA`
3. `corepack enable pnpm`
4. `./scripts/setup.sh`
5. Export `AMA_LLM_API_KEY`, `AMA_LLM_BASE_URL`, and `AMA_EMBEDDING_URL` in your shell, pointing to your OpenAI-compatible endpoints.
6. Open `SKILL.md` and map its retrieve, capture, synthesize, inspect, and delete rules onto your harness tool interface, using the six operations in `tools/`.

### Prompt to paste into your agent

```text
In the AMA repo at <path-to-AMA>, read SKILL.md and list each rule about when to retrieve prior context, capture turns, synthesize episodes, inspect state, and honor deletion requests. Expected output: a bulleted list of those rules with the heading each came from. Verify that every rule you list appears in SKILL.md text and none is invented.
```

### Caveat

The README describes the six operations and the policy only in general terms, so you must read the tools/ folder to see their exact interfaces before binding them.

Risk, credentials: The setup needs LLM and embedding API keys, which must be exported in your shell and never committed to the repo.

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/sherlockwz-ama/harness-agent-memory-skill/

## How to give an OpenClaw agent automatic recall and capture with AMA

Repo: [Sherlockwz/AMA](https://github.com/Sherlockwz/AMA) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README

Install the OpenClaw AMA plugin to recall and capture memory automatically each turn.

**Who it is for:** An OpenClaw user who wants the agent to remember past conversations without manual tool calls.

**The problem:** Each new OpenClaw session starts with no memory of earlier conversations or user preferences.

**The outcome:** OpenClaw runs the AMA plugin with automatic recall and capture, backed by a managed local SQLite and FAISS sidecar.

### Steps

1. `git clone https://github.com/Sherlockwz/AMA.git`
2. `cd AMA`
3. `corepack enable pnpm`
4. `./scripts/setup.sh`
5. `openclaw plugins install --link .`
6. `openclaw plugins enable openclaw-ama`
7. `openclaw ama doctor`
8. Add the `openclaw-ama` entry from the README's minimal configuration to your OpenClaw config, with `autoRecall` and `autoCapture` set to true.

### Prompt to paste into your agent

```text
In the AMA repo at <path-to-AMA>, run `openclaw ama doctor` and report each check it prints as pass or fail. Expected output: the list of checks with their results. Verify that `sidecarAutoStart` is enabled in the plugin config before you report the sidecar as healthy.
```

### Caveat

The README does not say which OpenClaw version is required, so confirm your OpenClaw install supports plugins before starting.

Risk, credentials: The OpenClaw config references your AMA LLM API key and embedding URL through environment variables, so keep those values out of version control.

Risk, credentials: Automatic capture stores conversation content in the local data directory, so review what is recorded and secure that directory.

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/sherlockwz-ama/openclaw-memory-plugin/

## How to start the AMA memory sidecar and confirm it responds with AMA

Repo: [Sherlockwz/AMA](https://github.com/Sherlockwz/AMA) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Launch the local AMA HTTP sidecar and check its health endpoint before wiring it in.

**Who it is for:** A developer building a custom agent who needs a local memory service running before connecting a client.

**The problem:** You cannot tell whether the AMA sidecar is running correctly before adding it to your agent.

**The outcome:** The AMA sidecar runs on localhost and returns a health response you can check.

### Steps

1. `git clone https://github.com/Sherlockwz/AMA.git`
2. `cd AMA`
3. `corepack enable pnpm`
4. `./scripts/setup.sh`
5. `.venv/bin/python -m uvicorn python.server:app --host 127.0.0.1 --port 8321`
6. `curl http://127.0.0.1:8321/health`

### Prompt to paste into your agent

```text
In the AMA repo at <path-to-AMA>, start the sidecar with the uvicorn command from the README on port 8321, then call the /health endpoint. Expected output: the raw health response body. Verify that the server is bound to 127.0.0.1 and not a public interface before reporting success.
```

### Caveat

The README only shows a health check as the smoke test, so a successful response does not prove that memory retrieval or LLM calls work.

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/sherlockwz-ama/local-memory-sidecar-smoke-test/

## How to reproduce AMA's LoCoMo and LongMemEval results with AMA

Repo: [Sherlockwz/AMA](https://github.com/Sherlockwz/AMA) | Difficulty: Advanced | Time: 120 minutes | Basis: From the README

Run the research pipeline on LoCoMo and LongMemEval_s to check the reported scores.

**Who it is for:** A researcher or student who needs to verify or extend the AMA results for an ACL 2026 paper comparison.

**The problem:** Published memory benchmark scores are hard to trust without rerunning the original pipeline on the same datasets.

**The outcome:** You run the AMA evaluation entry points and produce scores you can compare against the README's reference numbers.

### Steps

1. `git clone https://github.com/Sherlockwz/AMA.git`
2. `cd AMA`
3. `corepack enable pnpm`
4. `./scripts/setup.sh`
5. Follow the reproduction guide at `docs/reproduction.md` to obtain the LoCoMo and LongMemEval_s datasets and place them where the guide says.
6. Run the evaluation entry points under `python/AdaptiveMemory/Core`, as described in the reproduction guide.

### Prompt to paste into your agent

```text
In the AMA repo at <path-to-AMA>, follow docs/reproduction.md to set up the LoCoMo dataset and run one evaluation entry point from python/AdaptiveMemory/Core. Expected output: an overall LLM Score for the LoCoMo run. Verify that the dataset is placed where the guide specifies before comparing your score to the README's 0.774 reference.
```

### Caveat

The benchmark datasets are not redistributed in the repo, so you must obtain them yourself before any evaluation can run.

Risk, credentials: Full benchmark runs make many LLM and embedding calls, so the API keys in your environment will incur usage costs.

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/sherlockwz-ama/reproduce-am-paper-benchmarks/