# AI use cases for Flameox (morluto/flameox)

Repo: [morluto/flameox](https://github.com/morluto/flameox) | Use cases: 3 | Page: https://dedupedai.com/use-cases/morluto-flameox/

Flameox is a local tool that turns profiler, benchmark, and trace output into bounded runtime evidence that coding agents can use.

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

- Profile a Python benchmark for runtime hotspots: https://dedupedai.com/use-cases/morluto-flameox/profile-python-benchmark-hotspots/
- Analyze an existing profiler artifact without rerunning it: https://dedupedai.com/use-cases/morluto-flameox/analyze-existing-profiler-artifact/
- Connect a coding agent to Flameox runtime evidence: https://dedupedai.com/use-cases/morluto-flameox/give-coding-agent-runtime-evidence-via-mcp/

## How to profile a Python benchmark for runtime hotspots with Flameox

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

Capture a benchmark run and get bounded hotspot evidence without a project setup.

**Who it is for:** A developer who has a slow Python script or benchmark and wants to see where time goes before changing code.

**The problem:** Profiling output is hard to read and easy to over-interpret, and a manual profiling setup takes time.

**The outcome:** A bounded evidence result for the run that points to hotspots, with its limitations stated, ready for a hypothesis to be tested.

### Steps

1. Open a terminal in the project folder and note its absolute path with `pwd`.
2. Run `uvx flameox capture --provider direct --cwd "$PWD" -- python benchmark.py` with your own benchmark script in place of benchmark.py.
3. Read the returned evidence and any `next_page` value; replay a `next_page` continuation using its named analysis tool and arguments unchanged.
4. Treat the hotspots as exploration leads only; plan a separate experiment before claiming any speedup.

### Prompt to paste into your agent

```text
Using the flameox repo, capture a profile of <benchmark command> run from the absolute working directory <project path> with `uvx flameox capture --provider direct --cwd <project path> -- <benchmark command>`. Return the bounded hotspot evidence and its stated limitations. Verify that the result states what was observed versus inferred and does not claim causality.
```

### Caveat

Flameox records what the profile shows and does not establish causality or performance improvement, so any fix still needs its own measured experiment.

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/morluto-flameox/profile-python-benchmark-hotspots/

## How to analyze an existing profiler artifact without rerunning it with Flameox

Repo: [morluto/flameox](https://github.com/morluto/flameox) | Difficulty: Intermediate | Time: 15 minutes | Basis: From the README

Turn a saved profiler file into bounded evidence without running the workload again.

**Who it is for:** An engineer who already has a profiler or trace file from an earlier run and needs a structured read of it.

**The problem:** Rerunning a slow or hard-to-reproduce workload just to read its profile wastes time and may change the result.

**The outcome:** Bounded evidence extracted from the exact artifact file, with its provenance and limits recorded.

### Steps

1. Find the exact absolute path to your saved artifact file, such as an artifact.json from an earlier run.
2. Run `uvx flameox analyze preview_artifact /absolute/path/to/artifact.json` with that path.
3. Review the returned evidence, noting the observed inputs and listed limitations.
4. Keep the analysis_id only for the current session, since analysis IDs expire when evicted or when the server stops.

### Prompt to paste into your agent

```text
Using the flameox repo, run `uvx flameox analyze preview_artifact <absolute artifact path>` on the file at <absolute artifact path>. Report the typed evidence and coverage, separating observed from derived claims. Verify that the artifact path you passed is the exact file and that no parent directory was searched.
```

### Caveat

Flameox only reads the exact artifact path given and does not search parent directories, so the path must be correct and absolute.

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/morluto-flameox/analyze-existing-profiler-artifact/

## How to connect a coding agent to Flameox runtime evidence with Flameox

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

Register the Flameox MCP server so an agent can request profiles and evidence itself.

**Who it is for:** A developer using a coding agent such as Codex who wants the agent to gather runtime evidence on its own hypotheses.

**The problem:** Agents guess at performance causes without measured runtime data, and wiring up profilers for each session is tedious.

**The outcome:** The agent can call named Flameox analysis and capture tools through MCP and reports bounded, traceable evidence for its hypotheses.

### Steps

1. Run `npx flameox@latest setup` and choose your client, for example `--client codex --yes` for automation.
2. Restart or reconnect the MCP client as the setup output reports.
3. Run `uvx flameox mcp inspect` to confirm the available tool catalog, or use `--tool TOOL_NAME` for a focused schema. (inferred, check the README)
4. Ask the agent to use the Flameox tools for a specific hypothesis, then check that the returned evidence lists its limitations.

### Prompt to paste into your agent

```text
Using the flameox repo's MCP server, capture runtime evidence for the hypothesis <hypothesis> on the target <target command> with its working directory <absolute path>. Return the bounded evidence, coverage, and limitations for each claim. Verify that each claim is labeled observed, derived, or inferred and that no causal or speedup claim is made without a separate experiment.
```

### Caveat

MCP work belongs to the live request, so no detached job survives a restart and long captures must finish while the session is open.

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/morluto-flameox/give-coding-agent-runtime-evidence-via-mcp/