Typed in
A plain value. Paste it in from a spreadsheet, import a CSV file, or let another app send it in.
FREE · nothing is called
Free and open source · v1 alpha · Runs on your computer
Iron, not
Iron, not clay
A spreadsheet that fills itself in. Each column can ask an AI, run a rule, call a website, or go and research the answer. It runs on your own computer, it shows you the price in dollars before you start, and it never asks you to buy credits.
- Local AI apps
- 10
- Credits to buy
- none
- Cost per row, local
- $0.00
- Rows per table
- 1,000,000+
FILLING IN “INDUSTRY”
SPENT SO FAR $0.41
Company
□ Domain
◉ Industry
◉ Staff
1Acme Hydraulicsacme.co
queuedIndustrial
·240
2Northwind Foodsnorthwind.com
queuedFood & bev
·1,100
3Cedar Labscedarlabs.io
queuedBiotech
·62
4Brik & Cobrik.no
queuedConstruction
·380
5Vela Freightvelafreight.com
queuedLogistics
·915
Credits
Stop paying per row
Every other tool works the same way. You buy credits. One credit buys one answer. When the credits run out, your table stops. The AI itself is cheap. The credits are not.
Ferrum shows you the exact price before a run starts, counting every row the run will touch rather than just the rows on your screen. Afterwards it shows you what it did not spend, and why.
Fill in “Industry” for 12,480 rows
Rows to fill in12,480
Not changed since last time9,214 skipped
Cheap AI first, expensive one only if unsureon
Most it can spend on one row$0.05
You will pay about$0.94
Start
Try 25 first
OBS. What you pay today
- Clay, Free
- 500 actions + 100 credits / mo, 200 rows
- Clay, Launch
- from $167/mo, 2,500 credits/mo
- Clay, Growth
- from $446/mo, 6,000 credits/mo
- Bitscale, Growth
- $349/mo, 15,000 credits
- Bitscale, Booster
- $799/mo, 50,000 credits
- Freckle, Build
- from $99/mo
- Ferrum
- no credits. You pay the AI company direct.
Prices taken from each company's own website on 31 July 2026. Links are under the table below.
Comparison
Side by side
Prices come from each company's own pricing page, read on 31 July 2026, and the links are under the table. Where a company does not publish a figure, the row says so.
| Property | Ferrum | Clay | Bitscale | Freckle |
|---|---|---|---|---|
| Ownership | ||||
| LicenceCan you read and change the code | Open source | Closed source, hosted by them | Closed source, hosted by them | Closed source, hosted by them |
| Where it runsWho holds your data | Your own computer. All your data sits in one file you can copy, back up or delete. It never leaves. | Their cloud | Their cloud | Their cloud |
| Software price | $0 | $0 free tier Launch from $167/mo Growth from $446/mo |
$0 free tier Growth $349/mo Booster $799/mo |
$0 free tier Build from $99/mo |
| What it costs to use | ||||
| What you pay per rowThe cost of filling in one cell | Nothing. The AI company bills your own account directly. Use a local AI and it costs $0.00, and nothing leaves your computer. | Credits and actions, both counted | Credits | Credits. 1 credit = 1 answer. |
| Free tier allowance | There is no free tier, because there is no paid tier. | 500 actions + 100 credits / mo | 200 credits | 250 credits |
| Paid allowanceCheapest paid plan | No limit | 15,000 actions + 2,500 credits / mo | 15,000 credits | Not published on the pricing page |
| Unused credits | There are none to lose. | Build up to 2× your monthly amount | Build up to 3× on Growth and Booster | Not published on the pricing page |
| Row limit | As many as your disk holds. Tested with 1,000,000. | 200 free / 50,000 Launch / unlimited Growth | 50,000 free / 100,000 paid, per table | Unlimited tables on all plans |
| Use your own AIOn your own account, at your own price | 10 local AI apps and 21 AI companies, listed by name. Any model they offer. | Uses its own AI access, charged as credits. | Uses its own AI access, charged as credits. | Uses its own supplier network, charged as credits. |
| Control over spending | ||||
| Filling new rows by itselfAs data arrives | Yes, per column, with a dollar ceiling you set. At the ceiling it pauses and tells you rather than running on. | Not published on the pricing page | Not published on the pricing page | Not published on the pricing page |
| Deciding which rows to pay forFiltering before you are charged | One check over the whole table before any paid column starts. Costs the same on 100 rows and 100,000. Or build a filter from dropdowns, 20 tests, nested and/or. | Not published on the pricing page | Not published on the pricing page | Not published on the pricing page |
| Asking the same thing twice | The saved answer is used again, from any table, for 30 days. Failed answers are never saved. | Not published on the pricing page | Not published on the pricing page | Not published on the pricing page |
| Rows that did not change | Skipped. Every filled cell remembers what it was made from. Add 300 rows to 12,000 and you pay for 300. | Not published on the pricing page | Not published on the pricing page | Not published on the pricing page |
| Columns that run for nothingWork done without asking an AI | Describe the job in one sentence. Ferrum writes the rule, you approve it once, then it runs on every row for $0.00. The approval is locked to that exact rule. | Formulas and AI columns, inside their app | Action columns, inside their app | Inside their app |
| Price shown before you start | A dollar amount before you press Start, over every row the run will touch. No published price means the run does not start. Over $25 you type the amount in yourself. | The credit cost of each action is published. | The credit cost of each action is published. | The credit cost of each answer is published. |
| Reality | ||||
| Works offline | Yes, with a local AI | No | No | No |
| Money you did not spendShown as a number, not hinted at | Added up in dollars and split by reason. Covers AI columns and research agents only, so the figure is lower than the truth. Rows with no published price are counted on their own, never as zero. | Not offered. Credits get used up, not saved. | Not offered. Credits get used up, not saved. | Not offered. Credits get used up, not saved. |
| Turn a paid column into a free one | Offered after about 40 answered rows. Ferrum writes a rule from your column's own answers and marks it on rows it never saw. Saved switched off until you approve it. | Not published | Not published | Not published |
| Spending limits you set | A dollar limit per run, per table, and per cell. Reaching one pauses rather than fails, so the rows already filled keep their answers. | The credit balance is the cap. | The credit balance is the cap. | The credit balance is the cap. |
| Support | No premium support, but Adnan is always available to work out bugs and make improvements. Reach him at linkedin.com/in/adnanmanna or adnan@theascension.company. | Included, tiered | Chat from Growth, an engineer on Enterprise | Included |
Read 31 July 2026 from clay.com/pricing, bitscale.ai/pricing and freckle.io/pricing. Prices change, so check them yourself before you decide. Ferrum is not connected to any of these companies. The table shows how each product is sold, not how good it is.
What a column can be
Eleven kinds of column
A column can do eleven jobs. Six of them cost nothing at all. Ferrum labels the ones that spend money everywhere they appear, because that is the number you need before you press Start.
A rule
Two jobs that never cost anything. Tidy a value, like splitting a full name or pulling the domain out of an email. Or decide which rows are worth paying for, before a paid column sees them.
FREE · $0.00 on every row, foreverRead a linked table
Every contact points at its company. Fill the industry in once on the company, and all two thousand contacts read the same answer for nothing. Paying per contact instead is the most expensive mistake this kind of tool invites.
FREE · read straight from the tableCount a linked table
A number about everything on the other side of the link. How many contacts a company has, what the deals add up to, when you last spoke to anyone there. Ferrum counts it rather than asking an AI to count it.
FREE · counted, not askedWait a while
Hold each row for a set time before the next column runs. For a service that will not take a fast run, or when you want to slow one down on purpose. Rows still run side by side, and the editor tells you the real wall-clock time first.
FREE · nothing is calledAsk the AI
One question per row, on the AI you pick. A cheap AI can answer first and pass only the hard rows to the expensive one.
PAID · one AI call per rowCall a website
Any web service you have a key for. Ready-made setups for HubSpot, Pipedrive, Attio, Instantly, Smartlead and lemlist fill in the form for you, and you can change any field.
PAID · charged by whoever you callAsk a connected app
Ferrum speaks the same connector standard the AI apps do, so a tool you have already connected can answer a column directly. Apps that run on your own computer are free and send nothing anywhere, which is the part a hosted tool cannot copy.
PAID · unless the app runs on your computerResearch agent
It searches the web and keeps digging until it has an answer, with a limit on how far it can go. For questions one prompt cannot answer.
PAID · you set a dollar cap per rowTry one, then the next
Put several columns in order and Ferrum works down the list, stopping the moment one gives an answer good enough to keep. You pay only for the steps it actually needed. A cheap AI can answer most rows and hand the rest to the expensive one. The steps are ordinary columns, so an email hunt, a phone hunt and an AI hunt are one feature and not three.
PAID · stops at the first good answerSend it to another table
Copy these rows into another table, turning a list into records. It shows you what it will add, change and skip before it writes anything, and it matches on a key, so running it twice updates the record instead of making a second one.
FREE · $0.00, it stays on your computerYou describe a rule in one sentence and Ferrum writes it. It waits for you to approve it, and the approval is locked to the exact rule you approved: change one character and it stops and asks you again. If you would rather not deal with a rule at all, there is a plain filter you build from dropdowns, with 20 tests and nested and/or.
What you do not pay for
The cheapest AI call is the one you never make
Other tools count what you spent. Ferrum also counts what it did not spend, and puts a dollar figure on it. There are 20 of these, and 12 are already working the first time you open it, before you change a single setting.
The other 8 are switches you turn on: trying ten rows first, letting a column fill itself, the per-run and per-table limits, a rule that skips rows, dropping duplicate rows, two AIs in one column, running a local AI, and turning a paid column free.
- Cost saving features
- 20
- On before you change anything
- 12
- Default limit per AI cell
- $0.05
- Rows lost when it stops
- 0
Before you press Start
You see the price before anything moves
01 · The price comes first
$0.94
over every row the run will touch
Worked out before the run starts, over the rows it will really touch rather than the rows on your screen. If an AI has no published price the run does not start at all. Ferrum will not guess with your money.
02 · Try ten rows first
10 rows
then a low, likely and high figure
Not sure? Run ten rows, spread across the whole table rather than taken off the top. Ferrum then forecasts the rest from what those ten actually cost. With fewer than three usable rows it refuses to forecast at all. The ten rows are kept.
03 · Big spends are typed, not clicked
$25
above this you type the amount
Over $25 the Start button stays dead until you type the figure in yourself. A large run to an outside service also makes you type the number of rows. Clicking through a big spend by accident is not possible.
04 · Columns that fill themselves, capped
you set it
dollars per firing
Switch a column on and it fills new rows as they arrive. Give it a dollar ceiling and it stops there and tells you, instead of running until you notice. An import of 100,000 rows is one run, not 100,000.
05 · Every timer starts switched off
off
until you turn it on
You can tell a table to run every hour. It is made switched off, it can never run more often than every five minutes, and each run can carry its own dollar limit. Filling in a form never starts spending.
Rows that never reach a paid column
Six reasons a row costs you nothing
Before a paid column touches a row, Ferrum has six chances to decide that row does not need it. Every skip is added up in dollars, so you can see what each one saved you.
| Why the row is skipped | What it costs | On by default |
|---|---|---|
| Nothing changed since last timeEvery filled cell remembers what it was made from. Add 300 leads to a table of 12,000 and you pay for 300. A column that reads no other column has nothing to compare, so it always runs. | $0.00 | Yes |
| Your rule said noOne check over the whole table in a single pass, before any paid column starts. It costs the same on 100 rows and on 100,000. Or build a filter from dropdowns instead, with 20 tests and nested and/or. | $0.00 | You set it |
| A column it needs is emptyIf your question names a column and that column is blank, the row is dropped before a single word is sent. Put a ? after the name to say you can live without it. | $0.00 | Yes |
| The answer was already savedSame AI, same question, same row: the saved answer is used again, from any table, for 30 days. A failed answer is never saved. "Nothing found" is saved on purpose, because asking again would find nothing again. | $0.00 | Yes |
| It is the same row twiceImports and incoming data can drop rows you already have. Fewer rows to fill in. | $0.00 | You set it |
| Running it again would not helpWhen a cell is blank, Ferrum tells you which of five reasons it was. Three of them do not change if you run it again, and it says so instead of letting you pay a second time for the same silence. | $0.00 | Yes |
Ceilings you cannot go past
There is a limit on every cell before you set one
- Most one cell may spend
- $0.05 by default
- Most one cell may spend searchingand one search per cell
- $0.003 by default
- Limit per run, per table, per column
- you set it
- What happens at a limitrows already filled keep their answers
- it pauses
- Times a limit or sign-in error is retried
- 0
- Times it moves you to a dearer AI on its own
- 0
A pause is not a failure. The rows that finished keep their answers, and carrying on picks up where it stopped instead of starting again. If Ferrum cannot read a price, or cannot read your limit, it refuses the cell rather than running it without one. And the expensive AI never runs by itself: a cheap column that is unsure keeps its cheap answer and waits for you to say the word.
It gets cheaper the longer you use it
The bill goes down as the table gets older
Cheap first, dear only if needed
A cheap AI answers first. Only the rows it was not sure about go to the expensive one, and only when you say so. The saving is counted at the expensive one's price, because that is the call you did not make.
$0.00 ON THE ROWS IT ANSWERSNo row costs anything
Ten apps, eight addresses, and every row free. Nothing leaves the computer, and Ferrum checks the address really is your own machine so a tampered settings file cannot quietly send your prompts elsewhere.
$0.00 PER ROWStop paying for the column entirely
Ferrum studies the answers your paid column already gave and writes a rule that does the same job for nothing. The section below is how that works.
$0.00 PER ROW FROM THEN ONOne thing to know about the total. The dollar figure adds up what it saved on AI columns and research agents. The savings on calls to other websites are real, but they are not in that number, so the figure Ferrum shows you is lower than the truth, never higher. Rows where no price is published are counted on their own, never as zero.
From paid to free
Let the column pay for itself, then stop paying
A paid column answers 40 rows. That is enough. Ferrum takes up to 300 of your column's own answers, keeps 40 out of every 100 back, and writes a rule from the rest.
Then it marks its own homework. The rule runs on the rows it was never shown, and you get the score on rows it has never seen. A rule that just memorised your answers, one line per row, is caught and thrown out.
You also see up to five rows where the rule and the AI disagreed, so you can decide who was right. Nothing is asked of the AI again while any of this happens: the rule is written once, by an AI that costs nothing.
The rule is saved switched off. Your column keeps using the paid AI until you approve it yourself. After that the column costs $0.00 a row for as long as you use it.
Can a rule replace “Industry”?
Answers studied300
Rule written from180
Marked on rows it never saw120
Agreed with your AI114 of 120
Rows for you to judge5
Cost from here on$0.00
Approve the rule
Keep paying
Use any AI you like
Use any AI you like
These are the apps and companies Ferrum can talk to, listed by name. A local one costs nothing per row and nothing leaves your computer. A paid one charges your own account at your own rate, and Ferrum adds nothing on top.
Zero per row
Ollama · LM Studio · llama.cpp · LocalAI · vLLM · Modular MAX · Jan · GPT4All · LiteLLM · AnythingLLM
Ferrum asks the app which models it has and lists them for you, so the list is never out of date. Ten apps over eight addresses, because two pairs of them share a port. Ferrum checks the address really is your own computer before it sends anything.
$0.00 PER ROW · NOTHING LEAVES YOUR COMPUTERYour account, your price
OpenRouter · OpenAI · Anthropic · Google Gemini · Mistral · DeepSeek · Groq · xAI · Together AI · Fireworks AI · Cerebras · Moonshot (Kimi) · Perplexity · DeepInfra · Nebius · Hyperbolic · Novita · SambaNova · Cohere · AI21 · Z.AI
About 300 models with live prices through OpenRouter, plus your own key for the rest. You save a key once and give it a name. Columns point to the name, so the key itself is never written into a column.
THEY CHARGE YOU, WE DO NOTWhat it will not do
What it will not do
Every product page tells you what it does. This is the other half, and this is the part that tells you if Ferrum is wrong for you.
No hosted version
You run it yourself. That means a computer that stays on, your own AI keys, and nobody to call at 2am. If you want somebody else to run it for you, buy one of the products in the table.
No credits, but no refunds either
If a run spends $40 on an AI, that $40 is gone. Ferrum can put back the values a run replaced. It cannot put back the money.
No shop of data suppliers
There is no list of data companies to click. Here, a data source is any web service you already have a key for. That is more work on your first column, and far less lock-in by your hundredth.
One copy, one team
Several people can share one copy, with roles and permissions. There is no fee per person, and no wall between one company and another, because this is one program on one computer.
It spends money the moment you point a column at an AI
There is no practice mode you can end up in by mistake. A dialog shows you the price first, every cell has a limit before you touch a setting, and you can put a hard dollar limit on a run, on a table, and on a column that fills itself. But once you press Start, real money moves.
It is a tool, not a service
No setup call, no account manager, no roadmap you can steer with a contract. The code is the manual, and the issue tracker is the help desk.
Get it
Clone it, then point it at a model
You need Node 24 and a computer you control. Your database is one file, and the file is yours.
# clone, install, run git clone https://github.com/athm793/ferrum.git cd ferrum && npm install npm start # the engine, on your computer npm run web:dev # open this in your browser
This is v1 alpha, 1.0.0-alpha.1. It does everything on this page
today, but the database schema and the HTTP API can still change between alpha releases, and an
upgrade may ask you to migrate or to start a fresh database. Copy your SQLite file before you
upgrade, and pin a commit if something you build depends on it.