# AI use cases for Agent Workflow Platform (primorLee/agent-workflow-platform) Repo: [primorLee/agent-workflow-platform](https://github.com/primorLee/agent-workflow-platform) | Use cases: 4 | Page: https://dedupedai.com/use-cases/primorlee-agent-workflow-platform/ Agent Workflow Platform is an open-source TypeScript starter stack with an Electron desktop app, a FastAPI control plane, Python and Go workers, and a standalone workflow runtime for building Agent products. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Verify a task round trip from control plane to worker: https://dedupedai.com/use-cases/primorlee-agent-workflow-platform/verify-model-to-worker-task-round-trip/ - Chat with a local model in the AWP desktop app: https://dedupedai.com/use-cases/primorlee-agent-workflow-platform/chat-with-local-model-in-awp-desktop/ - Point the desktop app at a remote model endpoint: https://dedupedai.com/use-cases/primorlee-agent-workflow-platform/connect-remote-compatible-model-endpoint/ - Add a durable task ledger with resume instructions: https://dedupedai.com/use-cases/primorlee-agent-workflow-platform/add-durable-task-ledger-to-any-repo/ ## How to verify a task round trip from control plane to worker with Agent Workflow Platform Repo: [primorLee/agent-workflow-platform](https://github.com/primorLee/agent-workflow-platform) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Start the local Compose stack and confirm a submitted task completes through the FastAPI control plane and Python worker. **Who it is for:** A developer evaluating AWP who needs proof that the backend-to-worker execution path works on their machine before adopting any part of it. **The problem:** Reading about a control plane and worker layer does not show whether the pieces actually connect, claim tasks, and reach a terminal state. **The outcome:** A submitted allow-listed task returns a result, and a backlog of tasks is shown to reach terminal states with claims respecting worker capacity. ### Steps 1. Clone the repository with `git clone https://github.com/primorLee/agent-workflow-platform.git` and enter it with `cd agent-workflow-platform`. 2. Start the stack with `docker compose -f deploy/local/docker-compose.local-dev.yml up -d --build`. 3. Wait for readiness with `python scripts/wait_for_http.py http://127.0.0.1:8100/v1/health/ready --timeout 120 --json-field database`. 4. Submit one task with `python scripts/submit_local_task.py --timeout 60` and read the printed result. 5. Run the backlog check with `python scripts/verify_task_lifecycle.py --count 10 --timeout 120`. 6. Stop the stack while keeping its state with `docker compose -f deploy/local/docker-compose.local-dev.yml down --remove-orphans`. ### Prompt to paste into your agent ```text In the agent-workflow-platform repository, start the Docker Compose local dev stack, wait for the health endpoint to report ready, and submit one task with scripts/submit_local_task.py. Report the printed result and confirm that scripts/verify_task_lifecycle.py finishes with every task in a terminal state. ``` ### Caveat The Compose path needs Docker and Python 3.12, and the README says the helpers discover the generated API key without printing it, so you do not need to handle that key yourself. 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/primorlee-agent-workflow-platform/verify-model-to-worker-task-round-trip/ ## How to chat with a local model in the AWP desktop app with Agent Workflow Platform Repo: [primorLee/agent-workflow-platform](https://github.com/primorLee/agent-workflow-platform) | Difficulty: Intermediate | Time: 60 minutes | Basis: From the README Connect the Electron desktop workbench to an OpenAI-compatible local endpoint and get streamed replies. **Who it is for:** A developer who has a local model running and wants a real desktop chat interface with streaming, history, and session resume around it. **The problem:** A bare model endpoint gives no desktop UI, streaming pipeline, or persisted conversation state for a product prototype. **The outcome:** A streamed model reply appears in the Electron app, and the model-native session can resume after a restart. ### Steps 1. Install desktop dependencies with `npm --prefix apps/desktop ci`. 2. On macOS or Linux, launch with `AWP_AGENT_MODEL=llama3.2 npm --prefix apps/desktop run openai-compatible:electron`. 3. On Windows PowerShell, set `$env:AWP_AGENT_MODEL='llama3.2'` and then run `npm --prefix apps/desktop run openai-compatible:electron`. 4. Send a message in the desktop window and confirm the streamed reply renders. 5. Restart the app and confirm the session resumes. ### Prompt to paste into your agent ```text In the agent-workflow-platform repository, install the desktop app dependencies and launch the OpenAI-compatible Electron path with AWP_AGENT_MODEL set to the model name served by my local endpoint. Report whether a streamed reply appears in the window, and confirm that the conversation is still there after a restart. ``` ### Caveat The README says the reference adapter needs an OpenAI-compatible Chat Completions endpoint, and the example model name llama3.2 is only an example you must replace with a model your endpoint actually serves. 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/primorlee-agent-workflow-platform/chat-with-local-model-in-awp-desktop/ ## How to point the desktop app at a remote model endpoint with Agent Workflow Platform Repo: [primorLee/agent-workflow-platform](https://github.com/primorLee/agent-workflow-platform) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README Use a hosted OpenAI-compatible endpoint with an API token and an explicit opt-in flag. **Who it is for:** A developer who wants the desktop workbench to use a remote compatible model service instead of a local one. **The problem:** A remote endpoint needs a token and an explicit opt-in, and mixing those up can send credentials to the wrong place. **The outcome:** The desktop app talks to the remote endpoint with the configured model, base URL, and token, and only after the opt-in is set. ### Steps 1. Install desktop dependencies with `npm --prefix apps/desktop ci`. 2. Set `AWP_AGENT_MODEL` to the model name your remote endpoint serves. 3. Set `AWP_AGENT_API_BASE_URL` to the endpoint base URL and `AWP_AGENT_API_TOKEN` to your token. 4. Set `AWP_AGENT_REMOTE_API_OPT_IN=1` to allow the remote call. 5. Run `npm --prefix apps/desktop run openai-compatible:electron` and send a test message. ### Prompt to paste into your agent ```text In the agent-workflow-platform repository, configure AWP_AGENT_MODEL, AWP_AGENT_API_BASE_URL, AWP_AGENT_API_TOKEN, and AWP_AGENT_REMOTE_API_OPT_IN=1 for the endpoint at , then launch the openai-compatible Electron path. Report whether one test message gets a streamed reply, and check that the token is read from the environment and is not written into the repository. ``` ### Caveat The README requires the exact opt-in value AWP_AGENT_REMOTE_API_OPT_IN=1 for remote endpoints, and it does not describe how the token is stored beyond the environment variable. Risk, credentials: This path sends an API token to a remote endpoint, so store the token only in your environment and never commit it to the repository. 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/primorlee-agent-workflow-platform/connect-remote-compatible-model-endpoint/ ## How to add a durable task ledger with resume instructions with Agent Workflow Platform Repo: [primorLee/agent-workflow-platform](https://github.com/primorLee/agent-workflow-platform) | Difficulty: Intermediate | Time: 15 minutes | Basis: From the README Track tasks, heartbeats, and checkpoints in a file-backed ledger that a new session can resume from. **Who it is for:** A developer or agent running long multi-step work in a repository who loses progress when a chat or process stops. **The problem:** Work stopped mid-task has no durable record, so the next session must rebuild context from memory. **The outcome:** A ledger in the ignored .agent-workflow/ directory holds the task, its checkpoint, and a rebuilt resume instruction that a new session can follow. ### Steps 1. Clone the repository with `git clone https://github.com/primorLee/agent-workflow-platform.git` and enter it with `cd agent-workflow-platform`. 2. Add a task with `python workflows/runtime/scheduler.py add "Create a deterministic smoke test" 2 --group reliability`. 3. Record progress with `python workflows/runtime/scheduler.py heartbeat`. 4. Save a checkpoint with `python workflows/runtime/scheduler.py checkpoint "Run the regression and attach its output"`. 5. Stop the process, then in a new terminal or agent session run `python workflows/runtime/guardian.py resume`. ### Prompt to paste into your agent ```text In the agent-workflow-platform repository, add a task to the scheduler, send a heartbeat, and save a checkpoint with the status of . Then run guardian.py resume in a new session and report the rebuilt resume instruction, confirming it comes from the durable state under .agent-workflow/. ``` ### Caveat The README says the scheduler does not spawn Agents or execute tasks, and Guardian does not restart a process, so you must run the work yourself and only use the ledger to record and resume it. 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/primorlee-agent-workflow-platform/add-durable-task-ledger-to-any-repo/