Metered per token

Pay per token.
Not per seat.

HiveMind reads and edits files, runs shell commands, and reasons about your codebase from the terminal. Every task starts on a cheap model and escalates only when it gets stuck — so you aren't paying top-tier rates for routine work.

curl -fsSL https://hivemind.bmtai.in/install.sh | bash

Installs to ~/.local/bin. Works on Intel and Apple silicon.

How a turn actually runs

Not a pipeline you configure — a loop that closes on itself, and prices itself every time it does.

One HiveMind turnYour prompt enters a loop: a cheap model answers with most prompt tokens served from cache, tools run in parallel, one anchored span is rewritten, and the turn's cost is settled and printed. Roughly four ten-thousandths of a cent per turn.cost this turn$0.000012$0.000021$0.000029$0.000038settled, not estimatedYour promptterminalCheap modelcache 92%Tools in parallelread · search · shellAnchored editone span
Every task starts on the cheap default. Prompt prefixes stay byte-stable so most input tokens bill at cache rates, older turns fold into a summary at 75% of the window, andedit_file rewrites only the span that changed.

One request, start to finish

We hold$0.000210
You pay$0.000038
You get back$0.000172

A conservative estimate is reserved before the request is sent, settled down to the real cost when it completes, and returned in full if anything fails — not just the unused part. The running total prints in your terminal after every turn.

Where the cost goes

Three mechanics do most of the work. None of them is a model choice.

+2

Anchored edits

edit_file replaces one old_string with one new_string, so changing a line costs tens of output tokens instead of re-emitting the file. Output is the priciest token class and is never cached — this is the largest single lever on a session.

cachedfull rate

Prefix cache

Tool schemas serialise in a deterministic order so the prompt prefix stays byte-stable turn to turn. Cache-hit prompt tokens bill at roughly a fiftieth of the miss rate, and the live hit rate prints next to every response.

3.63.93.84.13.9MiB, compressed — v2.0.0

Four megabytes

One static Rust binary across five targets. No Electron shell, no Node runtime, nonode_modules — curl, then run. It starts in the time a splash screen would have taken.

Get running

01

Install

One command, no source access and no build step. The binary lands on your PATH.

curl -fsSL hivemind.bmtai.in/install.sh | bash
02

Sign in

A code appears in your terminal. Approve it in the browser and you're linked.

hivemind auth login
03

Start working

Add funds from ₹10 and go. No API key of your own to create or rotate.

hivemind activate

Why we ask for your Google account

Signing in with Google gives HiveMind your email address, which we use only to identify your account and track your prepaid balance — nothing else. We don't post on your behalf, read your other Google data, or share your email with anyone besides our payment processor (Cashfree) for billing records. Full details in our Privacy Policy.