# AI use cases for System1 Agents (ThinkFlowLab/system1-agents) Repo: [ThinkFlowLab/system1-agents](https://github.com/ThinkFlowLab/system1-agents) | Use cases: 4 | Page: https://dedupedai.com/use-cases/thinkflowlab-system1-agents/ System1-Agents is a Python toolkit that gives AI agents System 1 decision models (Jev, Laya, Cua-S1) for browser, computer, game and robotics tasks. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Route support tickets to a queue in one call: https://dedupedai.com/use-cases/thinkflowlab-system1-agents/route-support-tickets-with-s1a/ - Hand single decisions to a System 1 agent from Claude Code: https://dedupedai.com/use-cases/thinkflowlab-system1-agents/delegate-page-and-game-decisions-from-claude-code/ - Compare Jev and a chat model on the same agent: https://dedupedai.com/use-cases/thinkflowlab-system1-agents/compare-jev-with-a-chat-model-on-blackjack/ - Build a new System 1 agent for your own task: https://dedupedai.com/use-cases/thinkflowlab-system1-agents/build-a-system1-agent-for-a-new-task/ ## How to route support tickets to a queue in one call with System1 Agents Repo: [ThinkFlowLab/system1-agents](https://github.com/ThinkFlowLab/system1-agents) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Send each ticket to one of five queues with a System 1 model and a probability. **Who it is for:** A support lead or ops engineer who gets a steady flow of tickets that must be sorted into logistics, payment, returns, account or human queues. **The problem:** Sorting tickets by hand is slow, and a full chat model call per ticket costs more and takes longer than a single decision. **The outcome:** Each ticket gets one queue choice with a probability and confidence, returned as a JSON object in about 400 ms. ### Steps 1. `git clone https://github.com/ThinkFlowLab/system1-agents && cd system1-agents` 2. `uv sync && cp .env.example .env`, then put a Jev key in `.env` as TYPESAFE_API_KEY or OPENROUTER_API_KEY 3. Run `uv run s1a decide --state '{"ticket": "I was charged twice for order 4411 and I want the second charge refunded."}' --option logistics="delivery problems" --option payment="billing and charges" --option returns="return requests" --option account="login and account access" --option human="anything else"` (inferred, check the README) 4. Read the printed JSON: `choice` is the queue, and the per-option probabilities and `confidence` show how sure the model is 5. Loop over your ticket export with the same command and send low-confidence tickets to the human queue (inferred, check the README) ### Prompt to paste into your agent ```text In the system1-agents repo, route each ticket in to one of logistics, payment, returns, account or human using `s1a decide`. Return one JSON line per ticket with the chosen queue and its probability. Verify that every output has a choice that matches one of the five queue names. ``` ### Caveat The README gives the ticket router result only as a benchmark and a Claude Code example; the exact decide flags for five queues are inferred and should be checked against docs/agents.md. Risk, credentials: The Jev key goes in a local .env file, so keep that file out of version control and out of shared logs. 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/thinkflowlab-system1-agents/route-support-tickets-with-s1a/ ## How to hand single decisions to a System 1 agent from Claude Code with System1 Agents Repo: [ThinkFlowLab/system1-agents](https://github.com/ThinkFlowLab/system1-agents) | Difficulty: Intermediate | Time: 15 minutes | Basis: From the README Install the s1a skill so Claude Code or Codex sends one-off choices to a fast model. **Who it is for:** A developer who uses Claude Code or Codex daily and wants routine page, game or selection decisions to run faster and cheaper. **The problem:** Every routine choice inside an agent session costs a full chat model call, which is slow and expensive for simple picks. **The outcome:** The host agent decides when to delegate, runs the right `s1a` command, and reports the chosen option with its probability. ### Steps 1. `npx skills add ThinkFlowLab/system1-agents` 2. `claude plugin marketplace add ThinkFlowLab/system1-agents && claude plugin install s1a@system1-agents` 3. Put a Jev key in the plugin or repo `.env` as described in docs/skills.md 4. Ask Claude Code or Codex for a one-off selection, such as routing a ticket or picking a move in a game 5. Check the reported choice and probability, and keep arithmetic, deduction and free text with the chat model ### Prompt to paste into your agent ```text Using the system1-agents skill in this session, decide the best option for with the s1a decide command. Report the chosen option, its probability and the command you ran. Verify that the reported choice is one of the options you supplied. ``` ### Caveat The README says the skill decides when to hand off a task, but it does not say which tasks the skill will accept, so check docs/skills.md before delegating live work. Risk, remote script: npx skills add and the plugin install pull code from the ThinkFlowLab GitHub repo, so review the skill files before you trust them in your environment. 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/thinkflowlab-system1-agents/delegate-page-and-game-decisions-from-claude-code/ ## How to compare Jev and a chat model on the same agent with System1 Agents Repo: [ThinkFlowLab/system1-agents](https://github.com/ThinkFlowLab/system1-agents) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Run one Blackjack agent with Jev and with the chat model, then compare score and time. **Who it is for:** A researcher or developer who wants to check the speed and cost claims for System 1 models on their own machine. **The problem:** The README's speed and cost numbers come from its own runs, so a team needs to measure the same agent on its own setup before relying on them. **The outcome:** Two job folders with scores per episode, one for each model, that can be compared side by side. ### Steps 1. `git clone https://github.com/ThinkFlowLab/system1-agents && cd system1-agents` 2. `uv sync && cp .env.example .env`, then add a Jev key to `.env` 3. `uv sync --extra blackjack` 4. `uv run s1a run blackjack --model jev --rethink off --episodes 20` 5. `uv run s1a run blackjack --model llm --rethink off --episodes 20` ### Prompt to paste into your agent ```text In the system1-agents repo, run the blackjack agent with --model jev and then with --model llm, both with --episodes 20 and --rethink off. Report the score and wall-clock time of each job folder. Verify that both runs completed 20 episodes before comparing them. ``` ### Caveat The README's Blackjack timing comes from its own runs, and a 20-episode run on your machine may show different numbers, so treat the comparison as a fresh measurement. 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/thinkflowlab-system1-agents/compare-jev-with-a-chat-model-on-blackjack/ ## How to build a new System 1 agent for your own task with System1 Agents Repo: [ThinkFlowLab/system1-agents](https://github.com/ThinkFlowLab/system1-agents) | Difficulty: Advanced | Time: 1 hour | Basis: From the README Use the builder skill to probe a task and write a System 1 agent module. **Who it is for:** An engineer with a repeated decision task in a browser, desktop, game or robot setting that no prebuilt agent covers. **The problem:** Writing a decision agent from scratch means guessing whether a fast decision model fits the task before any code exists. **The outcome:** A new module under s1a/agents/ ending in a frozen SPEC, after the builder has probed the task with hand-written decisions. ### Steps 1. `git clone https://github.com/ThinkFlowLab/system1-agents && cd system1-agents` 2. `uv sync && cp .env.example .env`, then add a Jev key to `.env` 3. Open Claude Code in this checkout and ask it to build a System 1 agent for your task 4. Let the builder run 8 to 12 hand-written probe decisions and check whether the task needs deduction or arithmetic 5. Run the new agent with `uv run s1a run --model jev` and check the score (inferred, check the README) ### Prompt to paste into your agent ```text In the system1-agents checkout, use the builder skill to create a System 1 agent for , ending in a frozen SPEC under s1a/agents/. Report the new module path, the probe results and the score from one run. Verify that the agent stops if the task needs deduction or arithmetic. ``` ### Caveat The README says the builder skill runs in Claude Code from this checkout, so other hosts are not covered by the documented build path. 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/thinkflowlab-system1-agents/build-a-system1-agent-for-a-new-task/