AhanaWire

AHANAWIRE / VERSION 3.0

AhanaWire 98.5% less traffic.
1 TB sends as about 15 GB.

That is the very first send on our measured test link, before any cache exists. On fast links the first send may not shrink; repeats still do. Repeats shrink to about 4.8 GB. At $0.09 per GB of egress, a $90 terabyte costs about $1.35 to send the first time and about $0.43 on repeats. The receiving edge brings back the full original, byte for byte.12

AhanaWire is compression software you run as AW Endpoints on the link you already have. It sends fewer bytes to lower egress and bandwidth cost, and the receiving AW Endpoint restores the original byte for byte.

Effective speedup: 66.8× more data over the same wire on the first send, 208.9× on repeats.3

“Effective” means more original data gets through per byte sent; the link itself does not get faster.

Results vary by data. Try it on yours →

Your infrastructure. Your cache. Your existing link.

Our goal: speed up the internet one device at a time, so existing networks carry more and fewer massive data centers are needed.

TO SCALE · 1 TB, SCALED FROM OUR MEASURED TEST1

Original1 TB
First trip≈15 GB
1.5% of the original · 98.5% less
Next trip≈4.8 GB
0.48% · 99.5% less

Bars are linear. First trip needs no cache; the next trip needs the chunks already at the receiving site. Sources & scope ↓

Effective speedup

≈66.8× more data over the same wire

First send, before any cache exists: every 1 TB of original data goes out as about 15 GB.32

≈208.9× on repeats

Once the far site already holds the data, 1 TB repeats as about 4.8 GB.3

“Effective” means more original data delivered per byte sent. The link itself does not get faster.

47.7× effective speedup finishing logs

Version 1 finishes repeated log text that many times sooner on a full link.4

6.7× effective speedup on a link ten times faster

The same log clock, on a full link ten times the speed.4

Every byte you move costs money. AhanaWire makes the bytes you move tiny. For high-volume data transfer, that is the job.

Lower the egress bill. Move more data over the same wire. Nothing is lost. It works on data you already send.

Your egress bill

$0.43

Egress fees are charged on data leaving your network or cloud. AhanaWire cuts data transfer costs by sending fewer bytes; the receiving AW Endpoint restores the original byte for byte.

A repeat terabyte costs $0.43 instead of $90.5

Smaller than the usual tools

150 bytes

Smaller than zstd, gzip, and lz4 on the same data, on the first copy and the next.6

Nothing lost

17,101

Every job restored. Every damaged frame refused.7

How it works

Two AW Endpoints, one at each end of the link you already have. Each one compresses what it sends and restores what it receives, byte for byte. Traffic goes both ways.

Two AW Endpoints sending packets both ways A full-size packet leaves AW Endpoint A, is compressed to a small packet, crosses the wire, and is restored to full size at AW Endpoint B. Another packet goes the other way, from B to A, the same way. AW Endpoint A compress · restore AW Endpoint B compress · restore your existing link A → B B → A full size in small on the wire full size out Two AW Endpoints sending packets both ways A full-size packet leaves AW Endpoint A, is compressed to a small packet, crosses the wire, and is restored to full size at AW Endpoint B. Another packet goes the other way, from B to A, the same way. AW Endpoint A compress · restore AW Endpoint B compress · restore your existing link A → B B → A full size in small on the wire full size out

Same 1 TB, same wire: time on the wire8

Full size
1 TB on the wire
AhanaWirecompress + send
≈15 GB on the wire

Effective speedup: 66.8× more data over the same wire on the first send, so the same 1 TB is on the wire for about 1/67 of the time. Repeats: 208.9×.2

A SHORT WALKTHROUGH

Install. Activate. Send less.

Install, send, then bring the original back.
Illustrated sequence—not a live installation.

AHANAWIRE / 2 AW ENDPOINTS (1 PAIR)

01 / INSTALL

A pair on each machine.

Start compress and expand together. Each site keeps its own persistent disk cache.

$ docker compose up -d --build
Run after receiving the edge binary.
AhanaWire: 2 AW Endpoints (1 pair) and customer-owned cachesThe source application sends original bytes to compress on port 19752. A first frame or next-copy reference crosses your link to expand on port 19754. The receiver restores the original using its local disk cache. Each site also has the opposite-direction edge.YOUR MACHINERECEIVING MACHINEYour applicationOriginal restoredCOMPRESS19752EXPAND19754YOUR EXISTING LINKAW Endpoints readyLocal disk cacheLocal disk cache+ expand for return traffic+ compress for return traffic31,334 B rebuilt · sha256 match ✓
PAYLOAD EGRESS / MIX31,334 B
REFERENCE SPEED / P991.3953 msLocal HTTP round trip, four clients

Your key comes from your account; the signed Linux binary is public.

The egress meter illustrates the measured four-file mix, not a single frame. The speed counter reveals a separate measured reference; it is not timing this animation. Controls pause or replay the sequence. Reduced-motion settings are respected.

Send a repeat terabyte for about $0.43 instead of $90. Every byte on the wire carries about 208.9× more of your original data.52

The first trip stores a densified frame. The next trip sends a short reference, because the other side already holds the chunks and rebuilds the original. The link speed has not changed; each byte on it carries more original data.

$88.65 savedStorage. A $90 terabyte kept as about 15 GB on the first trip. ≈66.8× more original data per byte sent.5
$89.57 savedEgress. Repeats go out as about 4.8 GB per terabyte. ≈208.9× more original data per byte sent.5
Every job restoredOriginal mixed text through one measured pair moved far more data than the wire carried. Throughput figures are in the footnote.9

Plans and prices

Pick a plan and continue to secure Stripe checkout. Every plan starts there, the Free plan and the Pro trial included. A card is kept on file to confirm who you are; the Free plan is never charged. Prices are in US dollars. Applicable sales tax is calculated at checkout from your billing address.

Free

$0

500 GB of data a month, hard cap. Above the cap, traffic passes through uncompressed. No overage charges.

2 AW Endpoints (1 pair) · up to 2 keys

Enterprise: over 1 PB a month, custom AW Endpoint and key limits, data volume and support terms. Quoted, from $2,500/mo. Apply to be an enterprise pilot partner.

Cancel any time from your account; the plan runs to the end of the paid period and is not prorated. Overage is billed monthly. The Pro trial needs a card and ends without charge if there is no card on file. After checkout you get an email to set your password, then you sign in and create your keys. Terms · Privacy. Already a customer? Sign in.

Who it's for

Bytes on the wire

Fewer bytes leave your network. That is a smaller egress line.

The bar is the bytes that travel for the same data. The number sits beside the bar.6

Effective speedup on repeats, by data shape

On repeats, each byte on the link carries far more of your original data. The bar is that multiple of the link’s own rate, by data shape.102

Frame math filling a link, not a change in the link’s rated speed.

Three versions

Version 3 is what ships to customers. The first copy is stored. The next copy is short.

Version 1.0

Compression on every byte

The smallest lossless frame for the message. This mix leaves as about 1.5% of its size. The same message sends that frame again.

Version 2.0

Built for faster links

Tuned for faster links. A protected frame still opens if one half of it is lost.11

Version 3.0

Cache on your disk

The first copy is the version 1 frame. Once that site holds the chunks, the next copy is the short frame, a few dozen bytes on each of these four files.12

How version 3.0 wires into your machine

It uses the link you already run. Your payload stays on your network.

Each machine runs one compress and one expand. They share a disk on that machine. The frame crosses the link you already have. AhanaAI issues the key once. The payload stays on your network.

Your application sends the original to compress on port 19752. Chunks stay on your disk. The first frame crosses your link to expand on port 19754 at the other machine, which opens the original and keeps the chunks. The next send is the short frame. Your machine Their machine Your application the original text Compress port 19752 Expand port 19754 Their application the original, opened your link Disk cache on this machine First trip is stored here. Later trips send the short frame. Disk cache on this machine Chunks stay here, so the short frame opens in full. One key from AhanaAI, placed on the edge. One cache per machine. A second customer gets a second key and does not share the disk. Docker or Kubernetes starts the pair.
Compress listens on 19752. Expand listens on 19754. Both mount the cache on that machine. The gold line is the link you already have. The return trip is the same pair the other way: their compress on 19752, your expand on 19754.
  1. AhanaAI gives you one API key. You put that key on the edge.
  2. Your application sends the original to compress on port 19752.
  3. The first frame crosses the link you already run. Their expand on port 19754 opens it.
  4. Each machine keeps the chunks on its own disk.
  5. The next copy is the short frame, because their disk already holds the chunks.

Measured speed of the short frame

Nothing was lost. Every measured job was restored.

Original throughput on one measured pair is the height of the bar. Figures are in the footnote.9

Every job restored; every damaged frame refused.7

Keep the frame. Open the original later.

Keep the small copy. Open the original later.13

The mix

1 TB kept as ≈15 GB

Opens as the full original, byte for byte.

The long line

1 TB kept as ≈7.6 MB

Best case: one long line repeated. It opens as the full line.

The handoff

About 90% to 99% smaller

Chat, handoffs, and logs each keep a small fraction of their original size.

Where this mix saves

The same savings idea, on the kinds of text in these files.

Illustrative egress per terabyte on the measured files.52

Logs and alarms

$89.40 kept on the first copy, $89.80 on the next, per terabyte. Plants, utilities, energy, water, historians, facilities, operations-event text, fleet and logistics exceptions, retail and hospitality operations, and municipal operations.

Mixed days

$88.82, then $89.74 per terabyte. Shared services, multi-site operations, health, finance, insurance, and archive text in this shape.

Handoffs

$85.52, then $88.09 per terabyte. Contact centers, agent platforms, case platforms, claims, care coordination, and matter work. The next copy keeps $7.17 more per terabyte than zstd-3 on this file.

Short turns and captions

$80.79, then $86.24 per terabyte. Captions, rundowns, schedules, and short operational chat. The next copy keeps $13.62 more per terabyte than zstd-3 on this file.

How we measured this

Every figure on this page comes from measured files and runs, scaled up where noted. Test dates, exact byte counts, and link speeds are in the footnotes below.142

Full results.

Questions

What is AhanaWire?

AhanaWire is a compression product that stores a densified first frame and sends a short next frame so the other side can open the original. The goal is to spend less on egress and storage and to carry more original text on the same link.

Is it lossless?

Every job in the published run restored, and every original in the published tables was restored. That is the measured scope.

How much does the published mix save?

About 98.5% less traffic on the first send and about 99.5% less on repeats: 1 TB goes out as about 15 GB, then about 4.8 GB, scaled from a measured 31,334-byte mix (469 bytes first, 150 bytes next). At an illustrative $0.09 per GB, a $90 terabyte costs about $1.35 to send the first time and about $0.43 on repeats. Results vary with data type and how often data repeats.12

How does it deploy?

One compress and one expand per machine, ports 19752 and 19754, cache on disk, a key issued by AhanaAI, and the payload on the customer network. The signed 3.0.6 install is at https://ahanaai.com/ahanawire/install/.

How do I start?

Pick a plan at https://ahanaai.com/ahanawire/#pricing and continue to Stripe checkout: Free, Starter, Pro with a 14-day free trial, or Scale, monthly or annual. After checkout, set your password from the emailed link, sign in, create a key, and follow the signed 3.0.6 install at https://ahanaai.com/ahanawire/install/. For Enterprise, apply to be an enterprise pilot partner.

Apply to be an enterprise pilot partner

For more than the Scale plan covers: over 1 PB a month, custom AW Endpoint and key limits, data volume and support terms. Enterprise is quoted, from $2,500/mo. AhanaAI is also looking for a few companies to test AhanaWire on real traffic and share usage data in return. Self-serve plans start at checkout. This opens an email to jeremiah@ahanaai.com. There is no form inbox on this website.

Partners test on real traffic and share usage data in return.

A first afternoon is the pair on two machines you already run: create a key in your account, put it in a mode-600 file, run the signed 3.0.6 install.sh, and send one original across the link you already have. Start with the installation guide. The signed bundle is at releases/3.0.6, with instructions and the k8s.yaml inside the verified package.

The payload stays on your network. AhanaAI issues a key, not custody of the data.

Footnotes

  1. Measured test: a four-file mix of repeated operational text, 31,334 bytes, through an AhanaWire pair on the default setting. The first trip was 469 bytes with no cache, the repeat a 150-byte reference, and every file restored byte for byte. Re-measured on AhanaWire 3.0.0 on Nov 19, 2026, about 11:50 AM HST; AhanaWire 3.0.3 gives the same sizes (checked Nov 20, 2026). The 1 TB figures are extrapolated from that test by the same ratios (469 / 31,334 and 150 / 31,334). Dollar figures illustrate egress at $0.09 per GB. ↩
  2. Results vary with data type and how often data repeats. Start a free plan or the 14-day Pro trial to see your own numbers, or compare plans. ↩
  3. “Effective speedup” means more original data delivered per byte sent. The link itself does not get faster; each byte on it carries more of your data. Measured multiples: 66.8102 to 1 on the first copy and 208.8933 to 1 on the next copy, on the 31,334-byte mix, AhanaWire 3.0.0, Nov 19, 2026, default setting, 10 Mbps test link. At 1,000 Mbps the first copy is sent as 31,406 bytes (the original plus an 18-byte frame header per file), so the first-copy gain applies on slower links; the next copy is 150 bytes at every measured link speed, including 1,000 Mbps. ↩
  4. Log clocks: measured on repeated operational text, 47.7366 times sooner on a full 10 Mbps pipe and 6.6793 times sooner on a full 100 Mbps pipe. October 2026 in-process version 1 clocks, no added round-trip delay. Logs only. ↩
  5. Dollar figures are illustrative, at $0.09 per GB of egress, on the measured four-file mix: a raw terabyte is $90, the first trip saves $88.65 and the next trip saves $89.57 ($90 minus $89.57 is $0.43). The other shapes on this page (logs, mixed days, handoffs, short turns) are the measured files. ↩
  6. Bytes that travel for the 31,334-byte mix: AhanaWire next copy 150, first copy 469, zstd-3 814, gzip-6 960, lz4-1 1,088. zstd-3, gzip-6, and lz4-1 are CLI measurements on these same bytes (Nov 19, 2026; zstd 1.5.7, gzip 1.14, lz4 1.10.0). Each of those tools sends that frame again; AhanaWire’s next frame is the short reference once the receiving site holds the chunks. ↩
  7. 17,101 of 17,101 jobs restored and 16 of 16 damaged frames refused. The 17,101 count is the earlier published run on the October 2026 four-file mix; the AhanaWire 3.0.0 re-measure on Nov 19, 2026 restored every file identically and refused 16 of 16 damaged frames. Each bar is its own complete count, not one scale. ↩
  8. The How it works graphic is illustrative, scaled from the measured test (31,334 bytes sent as 469 on the first trip, AhanaWire 3.0.0, Nov 19, 2026). Time on the wire scales with the bytes sent. Compress and restore time at each AW Endpoint is not shown; it adds a little time at each end, and on a fast, uncongested link it can cancel out the time saved. The byte saving holds either way. ↩
  9. Lead measured pair: one pair, 16 clients, about 8 seconds, warm next frame, every job restored; memory stayed at 4.9 MB and 3.0 MB. On that pair the mixed day moved at 2,169.976 Mbit/s of original text while the wire carried 6.368 Mbit/s. The short frame added 2,163.608 Mbit/s to the pipeline. Logs moved at 1,983.256 Mbit/s onto 4.496 Mbit/s. Two further hosts restored the same 29, 46, 29, and 46 byte frames, every job. Second host: chat 21,926.04 jobs/s and 15.217 MB/s, logs 12,793.03 jobs/s and 163.751 MB/s. Third host: chat 14,007.81 jobs/s and 9.721 MB/s, logs 7,970.26 jobs/s and 102.019 MB/s.
    FileJobs/sOriginalWirep99
    Chat, 29 bytes70,972.2149.255 MB/s2.058 MB/s0.445 ms
    Handoff, 46 bytes48,388.01105.099 MB/s2.226 MB/s0.909 ms
    Logs, 29 bytes19,367.73247.907 MB/s0.562 MB/s1.277 ms
    Mixed, 46 bytes17,312.14271.247 MB/s0.796 MB/s1.548 ms
    ↩
  10. Pipeline figures are frame sizes filling a link of the stated speed: link speed times the original bytes, divided by the frame. Next frame, original text carried on a full 10 Mbps link: chat (29-byte frame) 239.310 Mbps, handoff (2,172 → 46) 472.174, logs (12,800 → 29) 4,413.793, mixed day (15,668 → 46) 3,406.087, the mix (31,334 → 150) 2,088.933; a 100 Mbps link carries ten times those (2,393.103; 4,721.739; 44,137.931; 34,060.870; 20,889.333). Added over the 10 Mbps link: 229.310, 462.174, 4,403.793, 3,396.087, and 2,078.933 Mbps. The first frame of the mix is 469 bytes; a full 10 Mbps pipeline of those carries 668.102 Mbps and adds 658.102 Mbps. ↩
  11. Version 2.0: at 10 Mbps the frames match version 1; at 100 Mbps this mix is 3,192 bytes, 9.8164 to 1. A protected handoff is 174 bytes and opens as 2,172. ↩
  12. Version 3.0 next-copy frames on the four files: chat 29, handoff 46, logs 29, and mixed day 46 bytes. ↩
  13. Archive sizes: the mix keeps 469 bytes and opens as 31,334 (a terabyte of that shape is kept as 14.967 GB). A repeated line of 1 MiB or 16 MiB is kept as 127 bytes (a terabyte of the 16 MiB line is kept as 7.569 MB). A handoff keeps 108 bytes and opens as 2,172; logs of 12,800 bytes are kept as 85; chat of 694 bytes as 71. The archive is the densified frame. Version 3 also keeps each chunk on a disk you run, so the next send is the short frame and that site opens the original while the chunks remain. At 1 MiB and at 16 MiB the 127-byte frame is smaller than zstd-3 (171 and 1,613), gzip-6 (3,152 and 48,890), and lz4-1 (4,205 and 66,119). ↩
  14. First-trip and next-trip sizes are from the AhanaWire 3.0.0 re-measure on Nov 19, 2026, on the same four-file mix, not a promise for every file. AhanaWire 3.0.3 gives the same sizes on this mix (469 bytes first trip, 150 bytes next trip, every file restored identically), checked in the 3.0.3 release build on Nov 20, 2026. The short frame, the pipeline, and the archive are in the October 2026 results note, with first-trip and next-trip sizes and the CLI baselines re-measured on AhanaWire 3.0.0 on Nov 19, 2026. Version 1.0 compresses every byte. Version 3.0 keeps a cache on a disk the customer runs. Every original in those tables was restored, and the 3.0.0 re-measure restored every file identically. Log-clock figures (47.7× and 6.7×) remain the October 2026 in-process version 1 clocks. The numbers on this page are those files and those runs. ↩