How does condense actually shrink the bill?

Your agent re-sends the whole conversation to the model on every turn, so by mid-session you are paying again and again for text the model has already seen. On our own traffic 67.7% of the bill is re-reading old conversation, and only 0.2% is new text you just typed. condense shortens that history on its way to the model. Helene 1 handles the newest turn, dropping the parts that are not pulling their weight and passing the rest through word for word, fast enough that you will not notice it. Adeline 1 goes back over older work that is already finished and rewrites it to about a tenth of its length. What always survives: your instructions, what you asked for, and the answers you got back. Code, patches and error messages are never touched.

Will condense ruin my cache hit rate?

No. condense is built around the cache rather than in spite of it. We split each request into three parts. The oldest part is shortened once, then locked and never touched again, so the provider keeps serving it from cache cheaply for the rest of your session. The middle is the part still settling, which is due to be rewritten anyway. The newest part is this turn's text, which Helene trims before it is ever cached at all. Because the locked part only ever grows at the end, shortening what comes after it cannot spoil what is already cached. Our measured cache hit rate is 98.6%, and the published benchmark prices every run from real provider bills, so a broken cache would show up as a bigger bill rather than hide.

How much will I actually save?

It scales with how long your sessions get, because the saving comes from the history an agent re-sends every turn. Our published live run removed 37.3% of the bill on chains averaging ~179k tokens; sessions peaking past 800k save 66–72%. Dollar-weighted across all our own spend it comes to 66%. Chains under roughly 4k are too short to be worth compressing. The benchmark is public, so you can run it on your own sessions before you pay us anything.

How does condense compare to headroom, caveman or RTK?

We published a head-to-head against headroom: minmax-bench replays the same real session through both, reads back the real provider bill, and prices both the same way. condense removed 37.3% of the bill against headroom's best mode at 13.8%, close to three times as much. Compared only over the shorter sessions headroom managed, condense is at 28.0%, still about twice their best. The difference is built in: headroom's default setting keeps its cache clean but never touches the piled-up history that is most of the bill, while its more aggressive mode spoils the cache and ends at 1.8%. RTK, caveman and ponytail are not really competitors. They go after the model's replies and tool output, which is a different part of the bill, and they work alongside condense rather than replacing it.

What counts as a token saved?

Only the context we remove. If condense strips 4,000 tokens from a request before it leaves for Anthropic or OpenAI, that is 4,000 tokens saved: tokens the provider never saw and never billed you for. Whatever passes through untouched costs you nothing here.

Do I still pay Anthropic or OpenAI?

Yes. condense sits in front of your provider and uses your own API key, so the provider still bills you directly, just for far fewer tokens. That is where the saving lands. What you pay us for doing the compacting is either a flat monthly tier or a metered rate on the tokens we save.

Every tier is a weekly pool of tokens saved. Free gives you 25M a week. Starter is $5/month for 500M, Pro is $20/month for 2.5B, Max is $50/month for 10B. The pool refills every week, and when it is empty you drop to Condense Lite, which keeps compressing at about 20%. You lose some of the savings, never the service, and nothing extra is charged. Calling the API directly instead of subscribing is metered on the same tokens saved: $0.90 per million for Adeline 1, $0.40 for Helene 1. See the full pricing page for team plans and how to size a tier.

What happens when my weekly pool runs out?

Compression does not stop. You drop to Condense Lite, the same Helene 1 and Adeline 1 running under a cap, saving about 20% of a prompt until your pool refills at the start of the next week. Lite is a state of the plan you are already on, not a downgrade to a cheaper one: every tier lands on it when its pool empties and every tier goes back to full strength when the pool refills. Short sessions are identical either way, because the cap never binds on them. The longer a session runs, the more it wants to compress and the wider the gap gets, which is the whole argument for a bigger pool.

Does condense store my prompts or my provider key?

Your upstream provider key is never stored. Only a sha256 fingerprint reaches the usage ledger, enough to attribute requests without holding the secret. We never train on your data. Content is deleted on a 7-day window, all storage and processing stays in the EU, and zero data retention is available on request. Read the security page for the full posture.

Blog

Updates from condense.chat

Weekly notes on what we shipped, what changed, and what we are building next. New posts every Saturday.

condense.chat is a context compression API and drop-in proxy for coding agents, RAG pipelines, and chat products. This blog tracks the work behind it — new releases of our Adeline compaction model, changes to the proxy and the OpenAI and Anthropic SDK routes, benchmark numbers, and field notes from running real workloads at scale. If you're new here, start with the first post, then read the API docs or the FAQ for how it all fits together.

August 13, 2026Release

Helene 1.1

Helene 1.1 beats bear-2 by 7.1 accuracy points and kompress-v2 by 12.5 at matched compression on the public Lost in Compression benchmark, across four answering models and all ten languages. Ask it to remove half your tokens and it removes 50.5%.

Read post

July 25, 2026Benchmark

Guilty Until Proven — grading agent trajectories

“No quality loss” is easy to claim on benchmarks that look nothing like real work. The quality half of minmax-bench replays long real sessions step by step, grades each trajectory against an uncompressed control, and reads intent over mimicry — condense holds quality while cutting 62% of tokens and 45% of cost. Open source: use it to debunk us and every other compaction method.

Read post

July 3, 2026Deep dive

The anatomy of a token bill

Where agents spend money, and how to minmax it. A real coding session, replayed with real models and priced cache-aware, comes in 72% under baseline. The rebuilt engine, the live public harness — bugs included — and the rerun commands are all public.

Read post

June 27, 2026Release

Introducing Helene 1

Helene 1, our newest compaction model, joins Adeline 1 in the condense family and is generally available today. On CoQA a model answering from its compacted context is more accurate than the same model reading the full, uncompressed transcript, while cutting almost a third of the tokens, ahead of every compressor we measured. And dense now drives Codex and OpenCode alongside Claude Code.

Read post

June 13, 2026Release

Built for teams. Building for the enterprise.

Organizations are live. One workspace, a shared credit pool, a per member budget, and a spend cap you set. The client installs once and then stays out of your way. Under the hood we have been laying enterprise groundwork: zero data retention, a security pass, and a SOC 2 report landing in about three weeks. We skipped a week to clean up after the hackathon.

Read post

May 31, 2026Field notes

30 teams, one weekend, 4 billion tokens

We sponsored basedhack, an invite-only AI hackathon in Vilnius, and handed every team a condense key. The proxy compacted 4B tokens of agent context over the weekend — up to $19k of provider spend the room never paid. Plus: Adeline now runs on our own GPUs, and condense joins the AWS Global Startup Program.

Read post

May 23, 2026Release

Adeline V1 ships. The API ships with it.

Adeline 1, Our compaction model, goes live today. And the proxy now speaks both the OpenAI and Anthropic SDKs — point your base_url at a provider route, keep your key, change nothing else. Includes the new rewrite mode and the numbers so far.

Read post

May 16, 2026Release

Hello, condense.chat

Faster context compaction for Claude Code and the Claude Agent SDK, in beta. Sign up, get approved, and run Claude Code through condense.chat on $5 of starter credit. Includes the cost chart, the dashboard preview, and the curl one-liner.

Read post