Harfi
Harfi is a browser-based prompt game: you write the instruction, a code parser counts the bullets and words, and the score is the same every time.
Harfi is the only prompt trainer I've seen that scores you with a parser instead of a language model, and that design choice is what makes it worth your time — the feedback is repeatable and unarguable, it names the rule that broke and by how much. Set 1 is free forever and trial 1.1 needs no account, so there is no reason not to run it and see how many strokes you take. If your AI output feeds a downstream parser, the JSON and transform sets map onto that exact failure mode. Pick a model-evaluated prompt course instead only if you want creative-feedback coaching.
Verified 4d ago · liveness 60/100 · cite: rightaichoice.com/tools/harfi
- Developers who need AI to emit parseable JSON, CSV or YAML
- Prompt engineers who want repeatable parser feedback instead of a model's opinion
- Technical writers drilling strict formatting rules like word caps and banned words
- Data teams who want a short daily drill on controlling model output
- Developers who want AI to write code for them — reach for GitHub Copilot or Cursor
- Teams needing collaboration, SSO, dashboards or an API to wire into existing tooling
- Buyers expecting an IDE extension or workflow integration
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Harfi if you want a curriculum, a certificate, or a tool that writes prompts for you — it grades the prompt you wrote, trial by trial, and nothing else.
Only Set 1 (8 trials) is free; the other four sets require the one-time $29 purchase.
In short
Harfi — Harfi is a browser-based prompt game: you write the instruction, a code parser counts the bullets and words, and the score is the same every time. Best for Developers who need AI to emit parseable JSON, CSV or YAML, Prompt engineers who want repeatable parser feedback instead of a model's opinion, Technical writers drilling strict formatting rules like word caps and banned words. Free to start; paid plans from $29.
What people actually say about Harfi — is it worth it?
We scanned public community sources for Harfi on Aug 19, 2026 and could not establish that the discussion we found is about this tool rather than something else sharing its name. Our own analysis of that scan says the posts were off-subject. Rather than publish a sentiment score built on the wrong subject, we publish nothing here and re-run the scan.
Viability Score
How well maintained and how widely used is Harfi? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Parser-based scoring in code, not language-model judgement
- 40 trials across 5 sets of 8
- Set 1 — shape: bullets, word caps, sentence counts, no preamble
- Set 2 — schema: clean JSON, CSV, YAML, tables, no stray fences
- Set 3 — forbidden words, banned letters, acrostics and lipograms
- Set 4 — transform: messy input, one exact right answer
- Set 5 — control: same prompt replayed against a second model
- Four medals per trial: par, birdie, ace, minimalist
- Minimalist medal for a prompt under 25 words
- Prompt length capped at 120 words or fewer
- Instant parser feedback naming the exact rule that failed
- Par is a target, not an attempt limit — unlimited retries
- Plays against Claude Haiku 4.5 for most trials
- A second model used in the reliability set
- Browser-based, no install required
About Harfi
Harfi turns prompt precision into a forty-trial game. Each trial hands you a specification the machine's answer has to satisfy — exactly three bullets, a hard sixty-word cap, valid JSON with no code fence, five words you may not use — and you write the instruction, in 120 words or fewer, that gets the model there. You never write the answer; you write the instruction that produces it. Shorter prompts score better, so the pressure is on compression as much as correctness. The trials are organised into five sets of eight: Set 1 covers shape (bullets, word caps, sentence counts, no preamble), Set 2 chases machine-readable output (clean JSON, CSV, YAML, tables, no stray fences), Set 3 is forbidden words, banned letters, acrostics and lipograms, Set 4 takes messy input and demands one exact right answer, and Set 5 replays the same prompt against a second model to see whether what worked on one survives on the other. Scoring is the differentiator: a parser in code counts bullets, counts words, parses the JSON and scans for banned letters. No language model grades you, so the same attempt always earns the same result and you can see exactly which rule broke and by how much. Every trial carries four medals — par, birdie, ace, and a minimalist clear with a prompt under twenty-five words — 160 in total. It runs in the browser, needs no install, and plays against Claude Haiku 4.5 for most trials. Developers, technical writers, and anyone piping AI output into a downstream parser are the natural audience.
Behind the Verdict
The appeal of Harfi is narrow and real: it makes prompt discipline a scored, repeatable exercise rather than a vibe check. Ask a language model whether your prompt was good and you get a different answer each time; ask Harfi's parser and you get words: 59 > 50 → FAIL. That difference matters most in the sets that mirror production work. Set 2 wants clean JSON, CSV, YAML and tables with no stray fences — exactly the failure that breaks a downstream ingest step. Set 4 takes messy input and demands one exact right answer, which is the transformation problem in miniature. Set 5 asks the question most prompt guides skip entirely: does the same prompt hold up on a second model? The 160-medal structure (par, birdie, ace, minimalist under twenty-five words) gives you a reason to come back after the first clear, and par being a target rather than an attempt limit means nothing locks and you can keep refining. The honest limits: the parser only checks whether your output satisfies the trial's stated rules — it does not grade creativity or overall prompt quality, so the feedback is precise but not broad. Set 1 is the free slice, and the paid side is a one-time unlock rather than an ongoing service, so there is no live dashboard or team workspace layer on top. There is no account needed for trial 1.1, and the trials play against Claude Haiku 4.5 for most of the run. It is a four-minute daily drill, not a curriculum, not an IDE tool, and not something you wire into a CI pipeline. If your job is getting a model to emit parseable output under hard constraints, that is exactly the muscle Harfi trains, and at $29 once to open all forty trials it is cheap against the rework it targets.
Researching Harfi? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Harfi actually fits — and what changes day-one when you adopt it.
Starts on free trial 1.1, works Set 1 without an account, then buys the unlock and moves through Set 2 — clean JSON, CSV, YAML and tables with no stray fences.
Outcome: Learns to state the exact output shape in 120 words or fewer and reads the parser line that names the broken rule, so the next prompt lands on the first stroke.
Runs Set 3 — forbidden words, banned letters, acrostics and lipograms — chasing the minimalist medal with prompts under twenty-five words.
Outcome: Builds the habit of checking every word in the instruction before sending it, which is the same discipline a hard word cap demands in production.
Sets the group on trial 1.1, which takes about four minutes and where most people need three strokes, then compares scorecards.
Outcome: A concrete, argument-free exercise: the parser output is the same for everyone, so the discussion is about the prompt rather than about the score.
Use Cases
- Learning to write instructions that get AI to emit JSON, CSV or YAML a downstream parser accepts without errors.
- Drilling prompt reliability on data transformation tasks, where messy input must produce exactly one correct output.
- Practising tighter, more constraint-aware prompts for code generation so you rework ambiguous instructions less.
- Testing whether a prompt that works on one model survives when replayed against a second one.
- A short daily four-minute drill on controlling AI output, one trial at a time.
- A team-building exercise that teaches prompt engineering through scored challenges.
Models Under the Hood
as of 2026-10-17
Limitations
- The game is scored by a code parser that counts bullets, counts words, parses JSON and scans for banned letters, so it evaluates only whether your prompt's output satisfies the trial's stated rules — not creativity and not overall prompt quality.
- You never write the answer; you write the instruction in 120 words or fewer.
- The trial set is the product: forty trials across five sets, with Set 1 the free slice and the remaining four sets unlocked by a one-time purchase.
- Nothing in the scraped material describes an API, plugin, IDE extension, or desktop or mobile app — it is a browser-based game.
as of 2026-10-04
Verification history
We have re-verified Harfi 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Harfi tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free
$0
Ideal for
Beginners curious about prompt control, or casual players who want to try a few trials without commitment. Also suitable for budget-conscious individuals wanting a taste of the game.
What this tier adds
The starting tier gives you Set 1 (8 trials) forever, your scorecard, and par/birdie medals. It's a complete introduction to the game's mechanics with no card required.
Plus
$29 once
Ideal for
Serious prompt engineers, developers who regularly use AI for machine-readable output, and teams wanting to train staff on precise prompt writing. Also for dedicated players who want the full challenge.
What this tier adds
Unlocks all 5 sets (40 trials), all 160 medals, and the full scorecard and rank. This is the full game, adding the Schema, Forbidden, Transform, and Control sets that cover advanced constraints and model reliability.
Setup time & first value
How long it actually takes to get something useful out of Harfi — broken out by persona, not the marketing-page minute.
Under a minute to first value: trial 1.1 runs in the browser with no install and no account, and the trial itself takes about four minutes, with most people needing three strokes. Making an account afterwards takes a signup form and exists so your scorecard survives across devices. The paid unlock applies immediately once purchased.
Switching to or from Harfi
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a general prompt-engineering course: keep the theory and move the practice to Harfi's parser-scored trials, where the feedback names the exact rule that broke.
- →From eyeballing AI output by hand: replace the manual check with Set 2's JSON, CSV, YAML and table specs, which fail loudly when the shape is wrong.
- →From a model-graded prompt evaluator: switch to Harfi's code parser, so the same attempt always earns the same result instead of a fresh opinion.
- ↗To an IDE assistant such as GitHub Copilot or Cursor if your goal is having AI write the code rather than writing better instructions for it.
- ↗To a full prompt-engineering curriculum if you want scheduled coursework and a certificate rather than a scored drill.
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Harfi”, and we withheld 6: 6 could not be judged, because “Harfi” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Harfi.
Official links
Tools that pair well with Harfi
Common stack mates teams adopt alongside Harfi, with the specific reason each pairing earns its keep.
Chatbox
Chatbox is a multi-model AI chat client with a local knowledge base and a Work Mode desktop agent that runs code and finishes tasks.
Claude
Claude is Anthropic's AI assistant for long-document analysis, agentic coding, and multi-step work inside one chat and Cowork surface.
Mistral
Mistral is the sovereign AI stack: frontier open-weight models you can self-host, fine-tune, or run in-region.
Featured Head-to-Head Comparisons
Harfi vs Roo Code
If you need a coding assistant that works today, choose Harfi—it interprets your intent and is available on a freemium model. Roo Code is intriguing with its multi-agent orchestration and local processing, but it's pre-launch, so only pick it if you're willing to wait and test unfinished software. For immediate productivity, Harfi wins.
Harfi vs Poolside Ai
If you're a solo developer or small team looking for a low-friction tool that understands your intent and spits out working code, Harfi's freemium model is the obvious pick. But if you're building high-stakes software in finance, healthcare, or defense, Poolside AI's open-weight models, on-prem deployment, and governance features are worth the enterprise investment. Choose based on your security and compliance needs, not just convenience.
Harfi vs Bito
If you're a solo dev or small team wrestling with AI codegen that often misses your intent, Harfi is worth a try. But if you're an enterprise already running Cursor or Claude Code across multiple repos and watching token bills climb, Bito's router and knowledge graph deliver measurable cost cuts (48% per task) without sacrificing quality — it's the safer bet for scale.
Alternatives to Harfi
View all
Chatbox
Chatbox is a multi-model AI chat client with a local knowledge base and a Work Mode desktop agent that runs code and finishes tasks.
Frequently Asked Questions
Categories
Best-of guides
Used Harfi? Help shape our editorial sentiment research.