Run dozens of coding agents in parallel without exhausting memory
Running multiple AI coding tools at once often strains your computer and freezes the screen. jcode addresses this by handling dozens of agents in parallel while significantly reducing memory usage and improving response times. Written in Rust, this tool creates a stable environment in the terminal for multiple AIs to work together.
Differences in memory and speed
jcode uses only about 3.3MB of memory per additional session. This is less than OpenCode by a factor of about 1.8, pi by about 24.1, Claude Code by about 51.3, and Codex CLI by about 58.8. The time to first input is also 19.1ms, which is faster than pi by a factor of 31.2, Codex CLI by 47.4, OpenCode by 54.9, and Claude Code by 183.9. This performance helps prevent system overload even in heavy multi-agent workflows.
Swarm feature that automatically coordinates agent conflicts
When multiple agents work in the same repository, if one agent modifies a file, other agents immediately become aware of the change. If the modified file is unrelated, they ignore it; if a conflict is suspected, they check the differences to resolve the issue. Agents can send messages to each other or announce to the whole group, facilitating collaboration. Additionally, agents can form their own teams to divide tasks, where the main agent acts as a coordinator and the generated agents act as workers.
Self-development mode that modifies itself
jcode provides a feature that goes beyond the limits of plugins or extensions, allowing agents to modify their own source code. When self-development mode is enabled, the agent edits its own source code, builds it, tests it, and then reloads the binary to continue working. This process happens automatically while the user maintains multiple sessions. This structure is specialized for jcode to improve and optimize itself.
Installation and supported services
On macOS and Linux, it can be easily installed using the curl command. It supports various providers, including Claude OAuth and OpenAI Codex OAuth. Token optimization, KV cache, and measurable metric improvement features are also provided.
For developers planning heavy parallel tasks
It is suitable for people using development workflows that require running multiple AI agents simultaneously. It is useful in environments where memory usage is sensitive or in complex projects where manually managing conflicts between agents is difficult. Before installing, you should check if your operating system is macOS or Linux and if the AI service provider you use is included in the support list. Also, when using self-development mode, be aware that the agent modifies its own code, and it is recommended to establish trust and a monitoring system for that process.
