TokenCode offers an open-source coding agent built for parallel runs
TokenCode is a terminal coding agent written in Go. The project says it can read and edit files, run commands, and search the web after a user gives it a task. It includes a `/race` mode where many agents try the same task separately. It also includes usage screens, token tracking, and conversation compaction.
Key points
- TokenCode is an open-source terminal coding agent written in Go.
- It can connect to Anthropic-compatible model services, with DeepSeek shown as the default setup.
- The `/race` mode lets up to 1000 agents work on the same task in separate workspaces.
- It filters empty changes and failed checks before using LLM scoring.
- It provides `/usage`, `/cost`, `/context`, and `/compact` commands for tracking and reducing token use.
Quick term guide
- coding agent
- An AI tool that writes or edits code from a person’s instructions.
- compaction
- The process where an AI summarizes its past conversation to save memory space.
- AI agents
- AI agents are AI tools that can carry out steps toward a goal, not just answer once.
- filtering
- A way to show only the items that match what you choose.
- LLM calls
- Requests sent to an AI language model like ChatGPT or Claude; more calls mean higher costs
- open-source
- Software whose code is shared publicly so others can inspect, use, or change it.
- compatible
- Parts that can work together without causing a mismatch.
- workspace
- A dedicated area inside the app where your project files are organized and connected