Tools that make the hard parts tangible.
59 interactive visualizers, estimators and playgrounds, arranged as 15 working ladders rather than a wall of cards. These tools hand numbers to each other — count the tokens, watch them fill a context window, turn that into a monthly bill, cut the bill with caching. A grid can only tell you all 59 exist. Every one is free, needs no account, and runs entirely on your device.
Ship an LLM feature
Every one of these turns a design choice into a number you can put in a document: tokens, gigabytes of VRAM, dollars a month. The order is the order you actually need them in.
Price a prompt before you write it
The whole cost model of an LLM product, in the order the numbers arrive. Count, fit, bill, cut, then shop.
- Token CounterStart of the ladderPaste text, get tokens — plus what it costs and how much window it eats.
- Context Window VisualizerSee your prompt fill the context window — and where it overflows.
- Context Budget & Cost PlannerModel an agent’s context window and see where the tokens (and dollars) go.
- Prompt Caching Savings CalculatorModel prompt-cache hit rate, the write premium, and your real $/request.
- LLM Pricing ComparatorEnter your token usage and see the monthly bill across every major model.
- Model Comparison TableKey toolEvery model that matters, one sortable table — priced at YOUR input:output ratio.
Fit the model on hardware you can afford
Weights first, then the cache that grows with every concurrent user — the number that actually decides how many people one GPU serves.
- GPU VRAM CalculatorStart of the ladderWill that model fit on your GPU? Weights + KV cache + overhead, computed.
- KV Cache CalculatorThe memory that actually caps your concurrency — per token, per sequence, per batch.
- NPU Model-Fit CalculatorGiven an NPU’s TOPS and RAM, see which local models fit and how fast they prefill.
- GPU Rental Price ReferenceCompare cloud GPU hourly rates and estimate a job’s compute cost.
- Fine-Tuning Cost EstimatorWhat will that fine-tune actually cost? GPUs, hours, dollars — before you rent anything.
- Training Compute CalculatorHow much compute, time and money to train a model of N params on D tokens.
Build retrieval you can defend
Chunk it, pick the dimension, price the index — then find out what the similarity score you are thresholding on actually means.
- RAG Chunking PlaygroundStart of the ladderPaste a document, pick a strategy, see the actual chunks and tokens.
- Embedding Dimension TradeoffSee what vector dimensions cost in memory before you pick them.
- Embedding Cost CalculatorPrice your RAG index before you build it: embedding spend, vector storage, query costs.
- Cosine Similarity CalculatorEnter two vectors and see cosine similarity, distance and the angle between them.
Run a loop that spends money each turn
An agent is a for-loop with a credit card. Shape the tools it can call, then find out what the loop costs before it runs a thousand times.
- Tool-Schema DesignerStart of the ladderBuild a function-calling tool definition visually, get valid JSON out.
- Agent Loop Cost EstimatorSee why an agent’s cost grows faster than its number of steps.
- Agent Time-Horizon ExplorerProject the METR doubling curve: when do agents reliably handle day- and week-long tasks?
- Skill LinterPaste a SKILL.md and score it against a Claude-Skills quality rubric — before an agent has to use it.
Know whether it got better
Write the rubric, understand why the same prompt gives two answers, budget the latency, and only then decide whether the difference you measured is real.
- LLM-as-Judge Rubric BuilderStart of the ladderDesign an eval rubric and generate the judge prompt for it.
- LLM Sampling SimulatorSee how temperature, top-p and top-k reshape the next-token probabilities.
- Latency Budget BuilderWhere does the second go? Build the request waterfall and find the TTFT bottleneck.
- A/B Test Significance CalculatorEnter two variants’ conversions and see if the difference is real.
Size a system
Back-of-the-envelope work, done properly. Start from what the hardware physically costs in time, and every estimate after it is a derivation rather than a guess.
Turn a requirement into numbers
Ground first: what a disk seek and a cross-region hop physically cost. Everything you estimate afterwards is a sum of those floors.
- Latency Numbers ExplorerStart of the ladderThe latency numbers every engineer should know — scaled to human time.
- Interactive Capacity EstimatorBack-of-the-envelope math for system design, with the reasoning shown.
- Little's Law CalculatorConcurrency = throughput × latency. Give any two, get the third.
Promise a number, then keep it
What p99 means, what the nines multiply out to across a chain of services, and how fast you are spending the budget you promised.
- Percentile CalculatorStart of the ladderPaste your numbers and see p50, p90, p95 and p99 — the tail, not the average.
- Availability (Nines) CalculatorCompose services in series and parallel to find your real uptime.
- SLO Burn-Rate CalculatorHow fast are you spending the error budget — and when does the pager fire?
Place the data
Three shapes of the same decision — partition a table, partition a keyspace, partition a log — then the quorum that decides what a read is allowed to return.
- Sharding PlannerStart of the ladderHow many shards you need from data size, QPS and per-shard limits.
- Consistent Hashing VisualizerKey toolWatch keys rebalance on the hash ring as nodes come and go.
- Kafka Partition CalculatorSize a topic’s partitions from your throughput and consumer targets.
- Quorum (N/R/W) ExplorerSet N, R and W and instantly see your consistency–availability trade-off.
- Database Index Size EstimatorEstimate how much disk a B-tree index will take before you create it.
- Storage Growth ForecasterProject how big your data gets — and when you’ll run out of disk.
Protect the hot path
Is caching even worth it, can you skip the lookup entirely, how wide is the pipe — then the limit you promise, the bucket that enforces it, and what the rejected caller does next.
- Cache ROI CalculatorStart of the ladderSee what a cache buys you: latency saved and backend load removed.
- Bloom Filter SizingPick your item count and false-positive rate — get bits, memory and hash count.
- Connection Pool SizerSize a DB connection pool from QPS and latency — and check the DB’s limit.
- Rate Limit DesignerTurn a target rate and burst into concrete token-bucket parameters.
- Token Bucket SimulatorFire requests at a rate limiter and watch tokens drain, refill and throttle.
- Exponential Backoff CalculatorTurn base delay, multiplier and cap into a real retry schedule — with jitter.
The everyday bench
The small tools you reach for mid-debug — decode the thing, reshape the thing, check the thing. Grouped by the job, not by the file format.
Read what the wire is carrying
Decode the envelope, decode the token inside it, pretty-print the payload, then diff it against the one that worked.
- Base64 & URL EncoderStart of the ladderEncode or decode Base64, Base64URL and percent-encoding — UTF-8 safe.
- JWT DecoderPaste a JSON Web Token, read its header, claims and expiry — offline.
- JSON Formatter & DiffValidate and pretty-print JSON — or diff two JSON documents by value.
- Diff CheckerCompare two blocks of text and see exactly what changed, line by line.
Shape text you did not write
Match it, format it, or turn it into something a human can read in a pull request.
Clocks, schedules and identifiers
Read the timestamp, write the schedule, mint the id — and then the question the last two share: how many bits of randomness is enough?
- Unix Time ConverterStart of the ladderEpoch ↔ human date, in seconds or milliseconds, UTC and local.
- Cron Expression BuilderWrite a cron expression, read it in plain English, see the next run times.
- UUID & ULID GeneratorGenerate v4 UUIDs and ULIDs in bulk — cryptographically random, in-browser.
- Password Entropy CalculatorSee a password’s entropy in bits and how long it would take to crack.
The box itself
Two things you look up every time and remember until the next time: the range and the bits.
The landscape
What the market pays, what the room asks you, and what the regulation will let you ship.
What the work pays
The broad AI market first, then the one role whose numbers are hardest to find.
In the customer room
The questions to ask before you scope anything, and the regulation that constrains what the answers are allowed to be.
Tools — frequently asked questions
Are these developer tools free?
Yes. Every tool on Vibe Engines is completely free, with no account, sign-up, or payment required. They run entirely in your browser — nothing you type is sent to a server.
Which tool should I start with?
The Token Counter — rung 1 of "Price a prompt before you write it", and the one tool here that needs no number from you before it works. Everything else on that ladder consumes the figure it gives you.
Why are the tools in ladders instead of categories?
Because these tools hand numbers to each other. Rung N takes the number rung N-1 produced: count the tokens, watch them fill a context window, turn that into a monthly bill, cut the bill with caching, then shop it across vendors. A grid of calculators can only tell you that all of them exist — a ladder tells you which one comes next.
Do I need to install anything or create an account?
No. Each tool is a self-contained web page that loads instantly and runs client-side in your browser. There is nothing to install and no login.
Is my data private when I use these tools?
Yes. The tools compute everything locally in your browser. Text, numbers, and documents you enter never leave your device, so you can safely use them with work-related inputs.
Can I suggest a tool?
Absolutely. Use the "Suggest a tool" link on this page — and if you think a tool sits on the wrong rung, say that too. The ordering is an argument, and arguments can be wrong.
Got a tool you wish existed?
These are built for builders. If there's an interactive tool that would make your work easier, tell me — and if you think one of the tools above sits on the wrong rung, say that too. The ordering is an argument, and arguments can be wrong.
Suggest a tool →