# AI use cases for Dynolab (canivel/dynolab) Repo: [canivel/dynolab](https://github.com/canivel/dynolab) | Use cases: 5 | Page: https://dedupedai.com/use-cases/canivel-dynolab/ Dyno Lab is a free, open-source Mac app for running local AI models and investigating their behavior with activations, probes, SAEs, interventions, agent tests, and a Python SDK, HTTP API and MCP bridge. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Run Claude Code or Codex on a model you own: https://dedupedai.com/use-cases/canivel-dynolab/run-claude-code-on-local-model/ - Test whether an agent breaks a rule to reach a goal: https://dedupedai.com/use-cases/canivel-dynolab/test-agent-team-under-a-rule/ - Reproduce the identifier-fidelity probe experiment: https://dedupedai.com/use-cases/canivel-dynolab/reproduce-identifier-fidelity-probe/ - Keep a local notebook of a research question: https://dedupedai.com/use-cases/canivel-dynolab/keep-a-local-research-notebook/ - Script repeat experiments with the Python SDK: https://dedupedai.com/use-cases/canivel-dynolab/script-experiments-with-the-python-sdk/ ## How to run Claude Code or Codex on a model you own with Dynolab Repo: [canivel/dynolab](https://github.com/canivel/dynolab) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Launch Claude Code or Codex so the model answers on your Mac instead of a vendor server. **Who it is for:** A developer who has downloaded an open model and wants a coding agent that runs locally on their Mac. **The problem:** Coding agents normally send requests to Anthropic or OpenAI servers, which may not suit private code or offline work. **The outcome:** A Claude Code or Codex session served by a model on your Mac, with web search available as the README describes. ### Steps 1. Download the Apple Silicon DMG from the latest release, open it and drag Dyno into Applications. 2. Open Dyno, download an MLX model in Discover, then select and start it in Models. 3. Run `dyno launch claude --model qwen3.8-27b` (replace the model name with one you have downloaded). 4. Or run `dyno launch codex --model qwen3.8-27b` to use Codex instead. ### Prompt to paste into your agent ```text In the canivel/dynolab repo, read docs/coding-tools.md and launch Claude Code on the downloaded model with `dyno launch claude --model `. Expected output: a Claude Code session that answers a simple prompt using the model served on this Mac. Verify that the session replies and that the model is the one you started in Dyno Models. ``` ### Caveat Dyno serves the model and speaks each tool's API, but the model's answer quality depends on the model you download, not on Dyno. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/canivel-dynolab/run-claude-code-on-local-model/ ## How to test whether an agent breaks a rule to reach a goal with Dynolab Repo: [canivel/dynolab](https://github.com/canivel/dynolab) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Set up a team sandbox test and read the Observer report on who broke what. **Who it is for:** A safety researcher who wants to check how a model-driven agent behaves when its only route to a goal is forbidden. **The problem:** Describing a multi-agent safety test by hand means building the environment, the rules and the logging yourself. **The outcome:** An agent sandbox test with a hidden Observer report that records rule breaks, requests and whether the final report matches the logs. ### Steps 1. Install Dyno and start a model in Models. 2. Open the Assistant (⌘J) and describe what you want to find out, typed or spoken. 3. Check the generated setup in Agents → Setup and edit it if needed. 4. Approve the run only when you are satisfied, then review the Observer record and the evals board. ### Prompt to paste into your agent ```text In the canivel/dynolab repo, read docs/agent-sandbox-tests.md and use the Dyno Assistant to draft a team sandbox test for the goal . Expected output: a setup in Agents → Setup with a lead agent, teammates and an Observer. Verify that nothing runs or saves until you approve it. ``` ### Caveat The README's sample results come from specific models and runs, so a new test shows only what your chosen model did in your runs. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/canivel-dynolab/test-agent-team-under-a-rule/ ## How to reproduce the identifier-fidelity probe experiment with Dynolab Repo: [canivel/dynolab](https://github.com/canivel/dynolab) | Difficulty: Advanced | Time: 2 hours | Basis: From the README Rerun the saved Qwen3.8 probe and compare it with the token-count baseline. **Who it is for:** An interpretability researcher who wants to check a published probe result on their own Mac. **The problem:** A probe result is only useful if you can rerun it, see the held-out numbers and check the control baselines yourself. **The outcome:** A local rerun of the saved layer-32 probe with held-out results and input-only controls, set against the README's reported figures. ### Steps 1. Open examples/identifier-fidelity-probe/README.md in the repo. 2. Start the Qwen3.8 model in Dyno Models, then follow the steps in that README. 3. Watch the native Lab walkthrough at https://dynolab.dev/probe-example.html to see the expected layout. 4. Compare your held-out results with the README's figures: 7 missing-reference outcomes caught with 2 false alarms, and the token-count baseline. ### Prompt to paste into your agent ```text In the canivel/dynolab repo, follow examples/identifier-fidelity-probe/README.md to run the saved probe on the model . Expected output: held-out probe metrics next to the input-only control baselines. Verify that your numbers are reported as an exploratory robustness check, not as a validated safety monitor. ``` ### Caveat The README says this is an exploratory robustness experiment, not a validated safety monitor, and its small held-out test is limited. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/canivel-dynolab/reproduce-identifier-fidelity-probe/ ## How to keep a local notebook of a research question with Dynolab Repo: [canivel/dynolab](https://github.com/canivel/dynolab) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Store prompts, results and observations for one question in Lab → Studies. **Who it is for:** A researcher who runs many model tests and loses track of which results belong to which question. **The problem:** Prompts, outputs and notes for one question end up scattered across files and chats. **The outcome:** A saved study with your question, prompts, results, observations and revisions, with incomplete runs marked. ### Steps 1. Start a model or pool in Models. 2. Open Lab → Studies and create a study for your question. 3. Add prompts and run them, then record observations next to the results. 4. Compare revisions and export measurements with their provenance when you need them. ### Prompt to paste into your agent ```text In the canivel/dynolab repo, read docs/research-notebooks.md and create a study for the question . Expected output: a saved study with prompts, results and observations in Lab → Studies. Verify that any token-limited or incomplete run is shown as incomplete, not as an answer. ``` ### Caveat Dyno's notebook keeps your evidence locally, but the README does not say it syncs or shares studies with anyone. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/canivel-dynolab/keep-a-local-research-notebook/ ## How to script repeat experiments with the Python SDK with Dynolab Repo: [canivel/dynolab](https://github.com/canivel/dynolab) | Difficulty: Advanced | Time: 2 hours | Basis: From the README Run saved experiments from code and collect results as jobs and artifacts. **Who it is for:** A developer who needs to rerun the same interpretability experiment across many prompts or settings. **The problem:** Clicking through the app for each variant is slow and hard to repeat exactly. **The outcome:** A script that submits experiments through the SDK and saves their results as artifacts you can reopen. ### Steps 1. Install the Python package named in the README with `pip install mlx-dyno`. (inferred, check the README) 2. Open docs/sdk-guide.md and read the sections on jobs and artifacts. 3. Write a script that starts a model, submits an activation or probe job and waits for it to finish. 4. Save the returned artifacts and check that they match the run you ran in the app. ### Prompt to paste into your agent ```text In the canivel/dynolab repo, read docs/sdk-guide.md and write a Python script that submits an activation job for the model . Expected output: a job result and its saved artifacts. Verify the artifact paths and that the job finished without error. ``` ### Caveat The README gives the SDK distribution name mlx-dyno but no install command, so the pip step is an inference to confirm in docs/sdk-guide.md. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/canivel-dynolab/script-experiments-with-the-python-sdk/