Less context, same Claude Code

condense.chat's harness intercepts your agent's context, passing it through our two compression models before it reaches the main model.

drop it in front of your coding agent, no key swap. Sign up for access.
why condense

You pay for the bloat.

capacity

Ship longer sessions.

typical saving−64%
cost

Shrink the bill.

typical saving−70%
quality

Same answers, fewer tokens.

faithfulness94.2%
vs the field

Best compression on the market.

We benchmarked condense against every context-compression approach we could run, on identical coding sessions. It removes more tokens, keeps more of the answer, and saves more money than all of them. Every metric, ahead. Read the full breakdown

2×

more tokens removed & money saved

same work, half the bill.

Almost no quality loss

the answer stays intact.

Read the engineering

Notes from behind the proxy.

How condense actually works, written up as we build it. Model internals, API decisions, and what each release changed. Longer than a changelog, shorter than a paper.

Release

Introducing Helene 1

Helene 1, our newest compaction model, joins Adeline 1 in the condense family and is live today. On CoQA it answers more accurately than the full uncompressed transcript while cutting almost a third of the tokens. And dense now drives Codex and OpenCode alongside Claude Code.

Read post
Release

Adeline 1 ships. The API ships with it.

Our compaction model goes live, 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.

Read post
Built by engineers shipping LLM infra at
Nord Security Kiloverse nexos.ai basedcollective_

Your next turn starts with fewer tokens.

Sign up to claim a key, then drop one command into your terminal.

Sign up →

Already have an account?