Condense

Shortens agent history before the model call and charges 15% of what that saves

Pricing model

15% of savings

Condense is a context-compression proxy for coding agents. You point requests at api.condense.chat instead of the provider, and it shortens conversation history the agent would otherwise resend on every turn. The Proxy API uses your own Anthropic or OpenAI key, so the provider still bills you, for fewer tokens. The fee is 15% of what that proxy saves, and there is no fee when it saves nothing. Two models do the compression: Helene 1.1 for everyday traffic at $0.40 per million tokens removed, and Adeline 1 for long agent traces at $0.90 per million tokens removed. Instructions, the request, and prior answers are kept. Code, patches, and error messages are left unchanged. API access is enabled for organization accounts after a call with the team. Personal accounts use the dense CLI.

What Condense does

Compress agent conversation historyReduce LLM input tokensProxy Anthropic and OpenAI requestsOne base URL in front of Anthropic or OpenAI, using your own provider keyProxy fee is 15% of measured savings, and zero when nothing is savedHelene 1.1 compresses everyday traffic at $0.40 per million tokens removedAdeline 1 compresses long agent traces at $0.90 per million tokens removedOldest context is shortened once and then left stable so the provider cache can keep hittingCode, patches, and error messages are not rewrittenProvider key is not stored; only a SHA-256 fingerprint is keptSpend caps per member and group, with savings reported by key

Pricing breakdown

Pricing model: 15% of savings

Frequently asked

Anthropic, OpenAI

Want a tailored answer?

See whether Condense fits your stack.

Techbible weighs Condense against what you already pay for, your team shape, and the work that's actually happening. Free to start.

Side by side

Compare Condense

Search the catalog or pick a similar tool to compare pricing, features, and fit.

More in AI Orchestration & MLOps

All tools →

CrewAI

Build and run autonomous multi-agent workflows using visual canvases or Python code

Arize AI

Monitor, trace, and evaluate large language model applications

PySpur

Build, test, and deploy AI agent workflows using code or visual drag-and-drop interfaces

AgentOps

Track, debug, and monitor AI agents and large language model interactions

Composio

Connect AI models and agents to external apps and developer tools

Traceloop

Track, evaluate, and monitor LLM application quality and latency in production environments

Voker

Track and evaluate conversational performance, user friction, and resolution rates across your AI agents

AgentX AI

Build, evaluate, monitor, and deploy multi-agent workflows across your enterprise infrastructure