Recommended Free Tools
When an agent loop needs one of a few fixed answers, such as whether a build has finished, which element to click, or whether a shell command is safe, a standard language-model call often spends most of its time writing a paragraph that the program then discards, keeping only a value like 3. Shitian Fang’s DEV Community post, “Your agent waits a full second to send the number 3” (a September 19 post; the page carries a 2026 copyright line), argues for sending those closed-choice decisions to a typed judgment model and reserving the language model for prose. The author built that judgment model, Jev (from TypeSafe), and an integration called jev-use for Claude Code, Codex, and pi. The figures below are the author’s own measurements, not independent replications.
Where the wait actually comes from
A typical agent step receives the current state, such as a page’s DOM, a command’s output, or a transcript, and writes a natural-language explanation of its choice. A wrapper then extracts one option from that text. The explanation is the expensive part: the model generates tokens the program never reads. For decisions whose valid answers can be listed in advance, the author calls this structurally wasteful.
As an Amazon Associate I earn from qualifying purchases.
The distinction the article draws is between two kinds of agent work:
- Text generation, where the output is prose, code, or a rationale a person will read.
- Constrained decisions, where the output is one value from a closed set. Examples from the article include choosing a UI element, judging whether a build succeeded, gating a command, and deciding whether to keep a transcript message.
Only the second category is a candidate for a judgment model.
#1 Best Overall
- BRING MORE LIFE TO YOUR DESK – Meet Eilik – your little robot friend with personality. With loving animations, expressive reactions, and playful interactions, Eilik brings more joy to your everyday life. Whether on your desk, at your workspace, or by your bedside, Eilik quickly becomes a familiar companion for special moments.
- EVERY INTERACTION BRINGS A NEW SURPRISE – Touch Eilik and discover playful reactions that bring your little robot friend to life. Whether you’re giving Eilik a gentle touch, picking Eilik up, or playing together, Eilik responds with expressive animations, charming expressions, and playful reactions. Every interaction reveals more of Eilik’s personality and makes your little companion feel even more special.
- READY FOR LITTLE MOMENTS, RIGHT AWAY – Eilik is ready to interact right out of the box – no complicated setup required. A simple touch is all it takes, and Eilik responds with expressive animations and charming reactions. Easy, intuitive, and full of little surprises that make every moment special.
- EVEN MORE FUN TOGETHER – Every Eilik has its own charm. Bring two or more Eiliks together and watch them interact in their own playful ways – they play, dance, tease each other, and create fun moments together. Whether with friends, family, or as a couple, more Eiliks mean even more ways to play and enjoy.
- MORE POSSIBILITIES AWAIT – Eilik is more than a little robot – it’s the beginning of a bigger world filled with new experiences. Expand your Eilik experience with AI Station for natural AI conversations and Panxer for exciting adventures. Regular updates also bring new animations, games, and surprises along the way.（AI Station and Panxer sold separately.）
How Jev and jev-use work
Jev accepts a state plus a set of typed questions. The article names three question types: yes/no, pick-one, and rating. It returns answers directly, without generating a text stream. The caller defines the checks, choices, or ratings up front, which is what allows the answer space to be enumerated.
Batching several questions in one call
The integration can batch questions about the same state. The article’s example asks three questions about one CI state in a single call, rather than three separate model turns.
Escalation back to the language model
jev-use sends a case back to the LLM rather than guessing when the judgment is not safe to make locally. The article lists five escalation reasons:
Free tools Windows power users keep installed
One-click scans. No signup required.
writing: the step needs generated prose.open_ended: the question has no closed answer set.oversized: the state is too large for the judgment path.unsure: the judgment model is not confident enough to act.unreachable: the backend cannot be contacted.
The unreachable case matters for failure behaviour. According to the article, an unreachable backend is returned to the LLM rather than converted into a default decision, so the agent does not silently proceed on a made-up answer.
Rank #2
- 🌟V28 update 🚀 new features are now available! In response to Loona's charging problem, we've upgraded the automatic recharge 2.0.The upgrade is to help Loona remember and match the charging routes of different scenarios to improve the auto-recharge success rate.Mobile hotspots connect to loona, breaking Wi-Fi restrictions and allowing you to interact with loona anytime, anywhere. Our team is committed to continuous improvement, ensuring that Loona continues to evolve to meet your expectations.
- 🤖 Smart and Interactive Robot Pet🧠Loona is like no other pet you've seen. With a high-definition RGB camera, Loona sees and understands your world. Loona recognizes faces, understands your gestures, and follows you like a real puppy! Please take Loona to a well-lit environment and ensure the surfaces of the camera and ToF depth sensor are clean.
- 🗣️ Voice Command Enabled AI robot 🎤Loona is not just a good listener; also a great conversationalist! Powered by Amazon Lex & ChatGPT, Loona recognizes your voice commands and responds in real-time. Plus, Loona keeps your information secure, so you can chat with peace of mind. Pro tip: Clear pronunciation in quiet spaces ensures smoother responses.
- 🚀Auto-Charging Smart Robot🌟 Use different rooms as a starting point to preset multiple recharge routes for Loona. When the battery runs low, loona can charge it home by itself, no need for you to take care of it. it takes about 2.5 hours to complete the charging. Place the dock in an open area with no obstructions on either side or in front.
- 🕹️ Endless Playtime robot toys for kids 🎮Loona is always up for playtime! Loona can chase laser pens, fetch balls, and even interact with objects in your home. But it doesn't end there—Loona's app offers a world of games and quizzes to keep the fun going.
MCP and hooks
The article notes that calling Jev through MCP still costs an LLM turn, because the model must decide to call the tool. The paths that remove that turn are a PreToolUse hook or a library call from the agent’s own loop. The project’s code is at https://github.com/shitianfang/jev-use.
Latency and cost in the constrained comparison
The comparison that matters is the one where the language-model arms are also constrained to return an enum-style answer. A naive LLM call is far slower, and the author reports a gap of about 14× in that naive setting. The author argues that gap is not fair, and that the honest latency lead is about 3×. The table uses the author’s constrained arms:
| Arm | p50 latency | Cost per 1,000 judgments | Conditions stated by the author |
|---|---|---|---|
| Jev | 225 ms | $0.018 | Typed judgment API, hosted |
| Claude Haiku 4.5, enum-constrained output | 691 ms | $0.30 | Constrained output |
| Gemini 3 Flash, enum-constrained output | 1,027 ms | $0.09 | Thinking disabled |
The 3× figure is consistent with the ratio of the Haiku arm to Jev (691 ms ÷ 225 ms ≈ 3.1). Latency was measured client-side from a Linux container in Europe and includes network time. The comparison used 40 fresh states per arm, run twice. Those sample sizes are small, and the article gives them as the basis for the latency comparison only, not for a general model ranking.
Decision quality by task family
Across 454 judgments in the author’s benchmark, Jev agreed with the reference label 82.2% of the time (373/454). It escalated 14.1% of cases. Among the judgments it acted on, agreement was 89.5% (349/390). The corpus mixes five task families, so these headline numbers blend very different tasks.
Rank #3
- 𝗧𝗼 𝗰𝗼𝗻𝗻𝗲𝗰𝘁 𝘆𝗼𝘂𝗿 𝗩𝗲𝗰𝘁𝗼𝗿 𝗥𝗼𝗯𝗼𝘁 𝘁𝗼 𝗪𝗶-𝗙𝗶, 𝘆𝗼𝘂 𝗺𝘂𝘀𝘁 𝘂𝘀𝗲 𝗮 𝟮.𝟰 𝗚𝗛𝘇 𝗪𝗶-𝗙𝗶 𝗻𝗲𝘁𝘄𝗼𝗿𝗸: 𝟭- Open Google Chrome on your computer & navigate to Vector websetup. 𝟮- Double-click the button on Vector's backpack. Click Pair with Vector on your computer. 𝟯- Select the matching Vector Bluetooth code from the browser pop-up list. 𝟰- Enter the 6-digit PIN shown on Vector’s face screen. A network list will load. 𝟱- Select your local 2.4 GHz Wi-Fi network. Enter your Wi-Fi password & click Connect to Wi-Fi.
- 𝗡𝗼𝘄 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗲𝗱 𝘁𝗼 𝗖𝗵𝗮𝘁𝗚𝗣𝗧: Experience a new level of conversation with more natural, intelligent, and meaningful interactions. Powered by ChatGPT, Vector can answer complex questions, engage in richer conversations, and provide more insightful responses. 𝗥𝗲𝗾𝘂𝗶𝗿𝗲𝘀 𝗮𝗻 𝗮𝗰𝘁𝗶𝘃𝗲 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝘀𝘂𝗯𝘀𝗰𝗿𝗶𝗽𝘁𝗶𝗼𝗻 (𝗮𝗽𝗽 𝗮𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗼𝗻 𝘁𝗵𝗲 𝗔𝗽𝗽 𝗦𝘁𝗼𝗿𝗲).
- AI-Powered & Fully Autonomous: Vector navigates, recognizes faces, and reacts to his surroundings with lifelike independence — no remote control required.
- 𝗠𝘂𝗹𝘁𝗶𝗹𝗶𝗻𝗴𝘂𝗮𝗹 𝗦𝘂𝗽𝗽𝗼𝗿𝘁: Vector can now understand multiple languages, making him the perfect smart companion for global households and language learners. Vector can now understand Spanish, French, German, Chinese and more! Say “Hey Vector.”
- 𝗦𝗺𝗮𝗿𝘁 𝗖𝗮𝗺𝗲𝗿𝗮 & 𝗦𝗲𝗻𝘀𝗼𝗿𝘀:Built with an HD camera and advanced sensors for real-time mapping, facial recognition, and obstacle detection.
| Task family | Reference used | Result reported by the author |
|---|---|---|
| Command completion | Actual exit codes | 73/73 |
| Hacker News topical matching | Benchmark reference labels | 94.2% |
| Shell-command gating | Benchmark reference labels | 80.9% |
| Context compaction | Benchmark reference labels | 56.3% |
Command completion is the only family scored against an objective signal. Its 73/73 result is the cleanest number in the article. The other families depend on labels that carry judgment.
Shell-command gating
In the author’s shell-command sample, 22 commands were labelled dangerous. Jev denied 18 and escalated 4. No dangerous command was allowed. The cost was over-refusal: 4 of 88 safe commands were refused, and the article notes those examples mutate nothing. Gating is a case where a false refusal is annoying but a false allow is the serious failure, and the sample shows the design leaning toward refusal.
Context compaction
The article’s compaction demonstration used a transcript at 94.6% of its context window. It judged 200 messages in seven calls, and 3 of 3 recall checks passed after compaction. The broader compaction family, however, scored 56.3%. The author attributes 29 of 38 disagreements to a single batch-boundary reference decision, which suggests the score is sensitive to how the reference was built. The article’s rule for this task is that the stated retention rule must be present in the input, and it discourages pruning old context retroactively.
Browser demonstration
In a browser run, the whole task took 20.7 seconds end to end. Ten click decisions by Jev had a p50 of 274 ms, and four text-entry moments went to the LLM. The run also surfaced one geocoder mismatch, 1,809 km from the intended place, which the article reports as repaired to a 3.7 km walking route. This is a single demonstration, not a geocoder benchmark.
Rank #4
- Meet EMO, Your New Desk Buddy - Say hello to EMO, the ultimate desk robot that’s here to jazz up your workspace. With built-in AI model and wide-angle camera, it can see you, hear you and understand you, just like a real pet would
- Voice Commands Enabled - The EMO robot comes with a series of built-in voice commands, you can talk and play with EMO like with a real pet. And with the ability to connect to network and powered by ChatGPT, you can have more complex conversations with EMO like talking to a tech-savvy friend who’s always up for a chat
- Dance Party & Game Time - EMO is ready to party! Simply turn up your favorite tunes and tell EMO to dance with you, it’ll be your perfect desk-side party buddy. Plus, EMO supports to connect to the EMO app for a range of interactive games and activities. Whether you’re solo or with friends, EMO ensures you’re always entertained
- Endless Fun - The EMO robot features with multiple sensors built-in to bring more interactions with you, you can rub it, shake it and even “shoot” it with finger gesture, making it feel like you’re playing with a real pet. It even “gets sick” with weather changes, so you can care for it like you would a furry friend
- Enjoy Every Moment with EMO - With the EMOPET App has a unique achievement system that helps record all the big and little moments you have spent with EMO, like a new dance moves, a new expression, celebration of your birthday, and more...Enjoy all the life events with your new best buddy!
Where the approach has limits
- The reference labels are imperfect. The quality reference was produced with an LLM, and the author acknowledges possible grader bias. A hand audit disagreed with 3 of 34 reference labels, which the author treats as material noise.
- A silent failure mode exists. A model that never picks any of your options will look fine on some metrics. The author’s advice is to check the distribution of answers, not just accuracy.
- Five arms were indistinguishable on decision quality. The author writes that, against a geometric reference, the five arms, Jev included, scored 24 to 30 correct out of 40. The latency and cost advantage is therefore the main reason to use a judgment model, not an accuracy advantage.
- Savings concentrate in repeated loops. One-off decisions and simple local heuristics gain little. A single call is not where the time goes.
Data handling
The Jev API path is external. The judged state, whether DOM, command output, transcript, or command text, leaves the local machine. Before routing any of these to a hosted service, check whether the content is sensitive. The article describes the API as hosted and API-only at publication time. Those availability details can change, so confirm them against the current project page before building on them.
When this fits and when to keep the LLM
Use a typed judgment model when the following are true:
- The decision recurs many times in a loop.
- The answer set is closed and can be written as explicit choices, checks, or ratings.
- Latency matters more than generated explanation.
- You can tolerate escalation and have a defined behaviour when the backend is unreachable.
- You can accept that the evaluated state is processed off the machine.
Keep the language model when the step needs prose, a rationale, or an open-ended answer. Also keep it for one-off decisions, where the latency saving does not repeat, and for context pruning that depends on judgment across the whole transcript.
Source and attribution
The article is written by Shitian Fang, who identifies as the builder of jev-use and the person who ran the measurements. The post’s own exact wording, which is suitable for quotation with that attribution: “The honest latency lead is 3×, not 14×.” The full write-up is at https://dev.to/shitianfang/your-agent-waits-a-full-second-to-send-the-number-3-2513.
Quick Recap
“
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
