Tokenade
A local token optimization tool for coding agents that reduces redundant token consumption in context through command output compression, code structure indexing, and documentation cleaning, and supports integration with agents such as Claude Code, Cursor, and Codex.
Timeline (last 90 days)
Tokenade was officially open-sourced in version 2.0 under a dual MIT OR Apache-2.0 license, with the code hosted on GitHub.
Tokenade is a token optimization tool for coding agents, and all processing is done on the local machine.
The author originally tried to run Tokenade as a paid product but ultimately chose to make it free and open source, because willingness to pay was misaligned: subscribers do not face token bills directly, while API developers who pay per token most need an agent-layer form, and the author did not want to run an agent business.
Tokenade provides command output compression, compressing noisy command output before the model reads the result, and compression can be bypassed with `tokenade raw <cmd>`.
Tokenade provides the `map`, `query <symbol>`, `skeleton <file>` and `semantic "<question>"` commands, letting agents obtain code structure information instead of whole files.
Tokenade provides a `read` command that converts PDFs, spreadsheets or long files to text, and the `web` and `search` commands do the same for web pages and search results.
Tokenade provides an `llm-proxy` local proxy that runs between the agent and the provider, collapsing tool results already present in history and sending identical tool outputs only once.
`tokenade install` automatically detects existing agents on the machine and completes integration, with no manual configuration required.
On Claude Code, for long sessions exceeding 200k tokens, Tokenade reduces token cost by an average of 39%.
Tokenade can be installed with two commands: `npm install -g @tokenade/cli` and `tokenade install`.
2 more timeline events
All Facts (12)
For short sessions or one-off tasks, Tokenade's token savings shrink noticeably, falling below the 39% average.
50%UnverifiedTokenade is a token optimization tool for coding agents, and all processing is done on the local machine.
50%UnverifiedThe author originally tried to run Tokenade as a paid product but ultimately chose to make it free and open source, because willingness to pay was misaligned: subscribers do not face token bills directly, while API developers who pay per token most need an agent-layer form, and the author did not want to run an agent business.
50%UnverifiedTokenade provides command output compression, compressing noisy command output before the model reads the result, and compression can be bypassed with `tokenade raw <cmd>`.
50%UnverifiedTokenade provides the `map`, `query <symbol>`, `skeleton <file>` and `semantic "<question>"` commands, letting agents obtain code structure information instead of whole files.
50%UnverifiedTokenade provides a `read` command that converts PDFs, spreadsheets or long files to text, and the `web` and `search` commands do the same for web pages and search results.
50%UnverifiedTokenade provides an `llm-proxy` local proxy that runs between the agent and the provider, collapsing tool results already present in history and sending identical tool outputs only once.
50%Unverified`tokenade install` automatically detects existing agents on the machine and completes integration, with no manual configuration required.
50%UnverifiedOn Claude Code, for long sessions exceeding 200k tokens, Tokenade reduces token cost by an average of 39%.
50%UnverifiedTokenade can be installed with two commands: `npm install -g @tokenade/cli` and `tokenade install`.
50%Unverified`tokenade dashboard` shows savings broken down by token category and model.
50%UnverifiedTokenade was officially open-sourced in version 2.0 under a dual MIT OR Apache-2.0 license, with the code hosted on GitHub.
50%