Launching Supercode beta - v0.1.56 Supercode is an open-source SWE agent native to your terminal, no cloud IDE, no web sandbox. Free to start with open models included (Deepseek, GLM, Minimax, Kimi and more). Bring your own API key for Claude, GPT, or any provider. Full machine access: read files, run commands, edit code, open apps, search the web. Granular permissions on your action. Persistent memory, and voice control.
No reviews yetBe the first to leave a review for Supercode
Maker
📌
I built Supercode because I wanted an SWE agent that actually lives where I work — the terminal. Most AI coding tools either lock you into a web UI or run in a cloud sandbox. That never made sense to me.
Supercode is open-source, runs natively in your terminal, and has full machine access — reads files, runs commands, edits code, opens apps, searches the web. Use free open models (Deepseek, Gemini) or bring your own API key for Claude/GPT. Every action needs your approval before it executes.
What's next: we're building a screen-actuation agent that sees your screen, hears your voice, and moves your cursor — plus MCP support, deeper app integrations, and more providers. Free to start. Would love feedback from the SWE community on what to prioritize.
Report
How does the granular permission system actually work in practice? Like, can I let it run npm commands freely but block anything that touches files outside the project directory?
Love that this lives right in the terminal without forcing me into a web IDE. One thing that would help me use it more confidently is a dry-run mode where I can preview the exact shell commands and file edits before they get executed, so I can sanity-check what the agent plans to do on a real codebase.
Report
Finally tried Supercode and the terminal-native setup feels right, no browser tab juggling. Having Deepseek and Kimi ready without an API key is a nice touch for quick experiments.
Report
running it straight from the terminal feels way more natural than spinning up another browser tab, and the granular permission prompts actually make me trust it with real repos
Report
the granular permissions stand out, that's the kind of thing most terminal agents skip and it makes actually handing over file and command access feel sane
Report
Been using the open models for a couple days and the terminal-native setup feels right, especially the granular permissions which gave me confidence to let it loose on bigger refactors.
How does the granular permission system actually work in practice? Like, can I let it run npm commands freely but block anything that touches files outside the project directory?
@rahimewfbb absolutely
Love that this lives right in the terminal without forcing me into a web IDE. One thing that would help me use it more confidently is a dry-run mode where I can preview the exact shell commands and file edits before they get executed, so I can sanity-check what the agent plans to do on a real codebase.
Finally tried Supercode and the terminal-native setup feels right, no browser tab juggling. Having Deepseek and Kimi ready without an API key is a nice touch for quick experiments.
running it straight from the terminal feels way more natural than spinning up another browser tab, and the granular permission prompts actually make me trust it with real repos
the granular permissions stand out, that's the kind of thing most terminal agents skip and it makes actually handing over file and command access feel sane
Been using the open models for a couple days and the terminal-native setup feels right, especially the granular permissions which gave me confidence to let it loose on bigger refactors.