1 · Start the local server
Run CodingSwarm AI on your machine or LAN host. Health check:
curl -s http://<CODINGSWARM_HOST>:8000/healthExpected: status=online
First 500 registrations free · Usage free
Slash
AI Coding Cost 75–85%
Reduce token usage. Lower AI coding costs — in Cursor and Antigravity, on projects of any size.
Smart rules for more efficient AI-assisted coding, powered by CodingSwarm AI.
- Target saving
- 75–85%
- Works with
- Cursor + Antigravity
- Project size
- Any
- Recurrent inference
- 1 ms
Illustrative example: 12,450 → 2,310 tokens (−81%).
- ▽ Smart FilteringRemove noise
- </> Context OptimizationKeep what matters
- ✦ Neuro-Symbolic InferenceVerify & certify
- ✓ Lower CostBetter outcomes
How it works
One setting. Starve the cloud meter.
Point Cursor or Antigravity at your local CodingSwarm server. Every request is filtered, optimized, and verified before it ever touches a metered cloud model.
-
1
Route your editor locally
Set the API base URL to
http://<your-host>:8000/v1— with modelcoding, or keep using the models you already like. Same models, same prompts, same workflow — only discounted through the swarm. -
2
CodingSwarm reasons, then proves it
Neuro-symbolic inference: neural models draft the code, symbolic checks verify it — only certified output is returned. A built-in discriminative model handles classification and regression with no training ahead of time: categories you provide are learned zero-shot at prediction time. Filtering strips noise and context optimization keeps what matters, deterministic at temperature
0.0. -
3
Cloud tokens only when needed
Unverified or out-of-scope work escalates to the cloud; everything else is solved locally. Fewer wasted retries, fewer bloated contexts — 75–85% fewer billed tokens.
Write better code. Use fewer tokens. Pay less.
Savings calculator
What does 75–85% off your AI coding bill look like?
Enter your current monthly spend on Cursor or Antigravity. The math is instant — no signup required.
You keep 75%$1,875stays in your pocket · monthly
You keep 81% (example)$2,025stays in your pocket · monthly
You keep 85%$2,125stays in your pocket · monthly
Yearly at 81%: $24,300 back in your pocket — per seat, on projects of any size.
Setup
Live in minutes, in Cursor or Antigravity
Three steps. Same local server, both editors.
Hardware requirement: Apple M4 or M5 (MacBook Pro / Max / Ultra) with minimum 128 GB RAM and 8 TB SSD — for your main node. CodingSwarm AI runs entirely on your machine, and every other device joins the swarm as a lightweight client.
2 · Verify the endpoint
From the machine running your editor:
cd swarm-client
./scripts/verify_codingswarm_io_endpoint.sh \
"http://<CODINGSWARM_HOST>:8000" \
"Write Python function fib(n)"Checks health + chat completion in one pass.
3 · Point your editor at it
- Base URL
http://<HOST>:8000/v1 - Model
coding - Temperature
0.0
- API base URL
http://<HOST>:8000/v1 - Model
coding - Max tokens
512 - Temperature / Top-p
0.0 / 0.9
Launch offer
First 500 registrations free.
Usage free.
No tiers, no token packs, no per-seat meter from us. Claim a spot, connect your editor, and start cutting your Cursor and Antigravity bills.
- ✓ Free registration (first 500)
- ✓ Free usage
- ✓ Works on any project size
Offer eligibility: the free launch allocation is limited to the first 500 eligible end-user registrations. Employees, contractors, and affiliates of competing AI coding-assistant providers — including Cursor (Anysphere) and Antigravity (Google) — are not eligible. CodingSwarm reserves the right to verify eligibility and to refuse or cancel any ineligible registration.
Why is it free?
No catch, and your code never leaves your machine. What CodingSwarm retrieves and learns from is orchestration — how coding requests are routed, verified, and distilled across models. That learning is what will allow full local neuro-symbolic orchestration later on: everything running on your hardware, cloud bill at zero. You get free usage today; only orchestration patterns are distilled — never your code.
The token
Use the swarm. Earn its value.
codingswarm.io is also a token. One side earns it, the other side needs it — and growth feeds both.
1 · Contribute to earn
Share your machine's orchestration with the swarm. The more distillation flows through you, the more token value accrues to you — everyday coding becomes earning.
2 · Demand grows with users
Every new builder beyond the first 500 must acquire token to register — so each wave of users adds demand for the token earlier users earned. A portion of every token payment is kept to fund development and upgrades for all customers, free and paid.
3 · Registration #501+ pays in token
The first 500 registrations are free. After that, registration is payable only in codingswarm token — a permanent demand sink built into the design.
FAQ
Reduce Cursor cost: questions, answered
How does CodingSwarm reduce my Cursor cost?
You point Cursor at your local CodingSwarm server instead of the cloud for every request. CodingSwarm filters noise, keeps only the context that matters, and returns verified code locally — metered cloud tokens are spent only when truly needed. Target: 75–85% fewer billed tokens.
Does it lower Antigravity costs too?
Yes. Antigravity connects through the same OpenAI-compatible API (base URL + model coding), so identical filtering and verification applies to Antigravity token usage.
Does it work on large projects?
Yes — project size doesn't matter. Filtering happens per request: only relevant context is forwarded, whether your repo has 100 files or 100,000. Savings grow with project size because bloated contexts are exactly what get trimmed.
What does “first 500 free, usage free” mean?
The first 500 registrations are free, and usage is free as well. Email us to claim one of the 500 spots — no credit card, no token packs.
Why is it free? What's the catch?
There is no catch on your code — it stays on your machine, always. CodingSwarm learns from orchestration: how requests are routed, verified, and distilled across models. That distilled orchestration knowledge is what will enable full local neuro-symbolic orchestration later. Your code is never sold, shared, or used to train anyone else's model.
Who is eligible for the free launch offer?
The offer is open to end users only, limited to the first 500 eligible registrations. Employees, contractors, and affiliates of competing AI coding-assistant providers — including Cursor (Anysphere) and Antigravity (Google) — are excluded. CodingSwarm may verify eligibility and refuse or cancel ineligible registrations.
Do I change how I work in Cursor or Antigravity?
No. You change one setting — the API base URL — and keep your prompts, keybindings, and workflow. CodingSwarm sits silently between your editor and the cloud.
Is CodingSwarm a cloud service?
No. CodingSwarm is a swarm of P2P learning nodes. Any machine on your local network can be shared into the swarm — coding and orchestration are then managed decentrally across your own hardware, with no central cloud and no single point of failure.
Does it need training on my codebase first?
No. Alongside neuro-symbolic inference, CodingSwarm runs a discriminative model for classification and regression that requires no training ahead of time — it learns zero-shot from the categories you provide at prediction time. It works on your code from the very first request.
How fast is local inference?
Blazing fast. Recurrent functions infer in about 1 ms — no network round-trip, no queue, no per-token wait. First-seen work takes longer; repeats are effectively instant.
How does the codingswarm token work?
Two flows, one loop. Contributors earn: the more orchestration distillation flows through your machine, the more token value accrues to you. Joiners pay: after the first 500 free registrations, new registrations are payable only in codingswarm token — so every wave of users adds demand for the token earlier users earned. A portion of every token payment is kept to fund development and upgrades for all customers, free and paid.
Where is my code stored?
On your machine. Your code never leaves your hardware or your local network — never uploaded to a cloud, never stored on our servers, never used to train anyone else's model. Only orchestration patterns are distilled to improve the swarm.