Condense is a context-compression proxy and CLI for Anthropic and OpenAI coding agents that reduces repeated context sent with model calls.
Why it matters
API Docs
API Available
overview
Condense is a context-compression proxy tool that enables developers and engineering teams to reduce repeated context sent by coding agents to Anthropic and OpenAI models. It manages and compresses agent context before API calls, with savings intended to increase during longer sessions. Users can run Claude Code, OpenAI Codex, or OpenCode through the dense CLI, or connect an existing agent by changing its base URL.
features
Condense operates as a proxy or through its dense CLI, compressing repeated agent context before calls to Anthropic or OpenAI. The service also provides spend and savings tracking and supports long-running coding-agent workflows.
use cases
Condense is intended for developers and engineering teams working with coding agents or building agents on Anthropic and OpenAI. Its documented integrations and use cases focus on reducing context and API costs during agent sessions.
how to use
Developers can use the dense CLI with supported coding agents or route an existing agent through Condense by changing its API base URL. Condense lists Anthropic and OpenAI as supported providers.
dense claude.pricing
Condense lists Free at $0, Starter at $5/month, Pro at $20/month, and Max at $50/month. Team Standard is $5 per seat per month and Team Premium is $20 per seat per month; Custom pricing is available by contacting sales. Separately listed token rates are $0.40 per 1 million tokens for Helene 1.1 and $0.90 per 1 million tokens for Adeline 1. Personal use is described as free, but the available plan data does not specify each tier’s included usage or limits. Teams integrating Condense into their products pay a share of the savings.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Similar Tools
Condense is positioned as a live proxy and CLI for compressing agent context. The alternatives described here differ in deployment approach or when they process context; their feature and pricing details may vary by configuration.
Provides an open-source Context Gateway that intercepts Claude Code and agent tool calls to run proactive background summarization and prune irrelevant tool definitions.
Offers similar proxy-level interception for coding agents, but advanced query-aware compression features rely on their external hosted engine rather than purely local compression.
Acts as a universal LLM gateway featuring native prompt compression helpers and caching layers that sit directly in front of Anthropic and OpenAI endpoints.
LiteLLM requires you to manage and configure your own proxy server or Python pipeline rather than executing a turnkey single CLI command like dense claude.
Uses a small local language model to calculate token perplexity and strip out low-information tokens from long prompts before dispatch.
It is a Python library rather than a transparent drop-in network proxy, requiring custom integration code and local GPU/CPU compute to compress prompts on the fly.
A CLI utility that bundles, cleans, and minimizes entire codebases into structured, token-optimized context files before passing them to LLM tools.
Operates ahead of time on static project files rather than acting as a live streaming proxy that trims continuous multi-turn chat history.
More on Stork
Other tools in this category, matched by shared tags