From eb707b01d18c7f1debee5c86a9e4198da3256335 Mon Sep 17 00:00:00 2001 From: Himanshu <102344906+coderwithsense@users.noreply.github.com> Date: Tue, 24 Feb 2026 17:48:22 +0530 Subject: [PATCH 1/3] Create CodingAgent.md --- GSoC-Ideas/2026/CodingAgent.md | 39 ++++++++++++++++++++++++++++++++++ 1 file changed, 39 insertions(+) create mode 100644 GSoC-Ideas/2026/CodingAgent.md diff --git a/GSoC-Ideas/2026/CodingAgent.md b/GSoC-Ideas/2026/CodingAgent.md new file mode 100644 index 0000000..5829198 --- /dev/null +++ b/GSoC-Ideas/2026/CodingAgent.md @@ -0,0 +1,39 @@ +# TODO: Idea Title + +## Background and Motivation +My motivation for building CodingAgent is that there is still no truly good open-source coding agent that works like the best closed-source tools such as Codex, Antigravity, or Claude Code. +I’ve used all of these extensively and I’m frustrated that I can’t plug my favorite open-source models into a comparable, powerful agent experience. +Key capabilities like robust persistent memory, solid MCP integration, and local vector-based storage for long-term context are either missing or poorly implemented in current open-source options. +I want to change that by creating an agent that treats memory and context as first-class citizens, not afterthoughts. +My goal is to build something that operates at the level of serious engineering workflows, not just toy demos. +In essence, I want to create the “clawdbot for coding” — a scalable, developer-first AI coding agent that is fully open, extensible, and production-ready. + +## Overview of Tasks +* Setup TypeScript CLI skeleton with commander + multi-LLM config +* Build agent loop: goal → plan → file/shell tools → memory compaction +* Test end-to-end: codingagi "create hello.ts" + +Candidates are expected to refine these tasks in their GSoC proposals. + +## Requirements + +* Model Agnostic - Works seamlessly with any LLM (Claude, OpenAI, Ollama, Kimi) via simple config, not locked to one provider +* Local vector store + structured notes that survive sessions and context limits (beats all open source agents) +* Reliable goal→plan→execute→observe cycle with MCP tools, file editing, and shell execution at Claude Code/Antigravity level (This point is very important to be explained in detail in the proposal, there are many complex plans in the market which are actually not suitable for this project, i want something straight forward, amazing and simple to use for the end user.) + +## Resources + +* Learn how memory layer is used here: https://github.com/opencode-ai/opencode +* excellent for CLI structure and local model integration: https://github.com/google-gemini/gemini-cli +* Full video tutorial building CLI agents with MCP, memory, and multi-agent workflows: https://www.youtube.com/watch?v=eNbza7cBL0U +* Loves Antigravity's autonomy/browser integration but warns of destructive ops, security gaps, and enterprise compliance risks vs Claude's safer design: https://blog.getbind.co/antigravity-vs-claude-code-which-is-better-for-developers/ + +## Mentors + +* GitHub: TODO; Discord: TODO +* GitHub: TODO; Discord: TODO +* Look for mentors with the role `@TODO` in our discord servers + +## Communication Channel + +Join our Discord servers (https://discord.gg/xnmAPS7zqB and https://discord.gg/fuuWX4AbJt) and discuss this idea in TODO. From 3d8f734f886f9a4f56667d5c92b48f999acdc538 Mon Sep 17 00:00:00 2001 From: Himanshu <102344906+coderwithsense@users.noreply.github.com> Date: Thu, 26 Feb 2026 17:55:50 +0530 Subject: [PATCH 2/3] Update CodingAgent.md --- GSoC-Ideas/2026/CodingAgent.md | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) diff --git a/GSoC-Ideas/2026/CodingAgent.md b/GSoC-Ideas/2026/CodingAgent.md index 5829198..2010912 100644 --- a/GSoC-Ideas/2026/CodingAgent.md +++ b/GSoC-Ideas/2026/CodingAgent.md @@ -1,12 +1,12 @@ -# TODO: Idea Title +# CodingAgent: Open-Source CLI AI Coding Assistant ## Background and Motivation -My motivation for building CodingAgent is that there is still no truly good open-source coding agent that works like the best closed-source tools such as Codex, Antigravity, or Claude Code. -I’ve used all of these extensively and I’m frustrated that I can’t plug my favorite open-source models into a comparable, powerful agent experience. +Our motivation for building CodingAgent is that there is still no truly good open-source coding agent that works like the best closed-source tools such as Codex, Antigravity, or Claude Code. +We’ve used all of these extensively and we are frustrated that I can’t plug my favorite open-source models into a comparable, powerful agent experience. Key capabilities like robust persistent memory, solid MCP integration, and local vector-based storage for long-term context are either missing or poorly implemented in current open-source options. -I want to change that by creating an agent that treats memory and context as first-class citizens, not afterthoughts. +We want to change that by creating an agent that treats memory and context as first-class citizens, not afterthoughts. My goal is to build something that operates at the level of serious engineering workflows, not just toy demos. -In essence, I want to create the “clawdbot for coding” — a scalable, developer-first AI coding agent that is fully open, extensible, and production-ready. +In essence, We want to create the “clawdbot for coding” — a scalable, developer-first AI coding agent that is fully open, extensible, and production-ready. ## Overview of Tasks * Setup TypeScript CLI skeleton with commander + multi-LLM config @@ -30,10 +30,8 @@ Candidates are expected to refine these tasks in their GSoC proposals. ## Mentors -* GitHub: TODO; Discord: TODO -* GitHub: TODO; Discord: TODO -* Look for mentors with the role `@TODO` in our discord servers +* GitHub: [coderwithsense](https://github.com/coderwithsense/); Discord: rmvfsx ## Communication Channel -Join our Discord servers (https://discord.gg/xnmAPS7zqB and https://discord.gg/fuuWX4AbJt) and discuss this idea in TODO. +Join our Discord servers (https://discord.gg/xnmAPS7zqB and https://discord.gg/fuuWX4AbJt) and discuss this idea in [general](https://discord.com/channels/1022871757289422898/1180497744666837003). From 35efe6dc65ca7fe13fab472e3699b9e426cba86c Mon Sep 17 00:00:00 2001 From: Himanshu <102344906+coderwithsense@users.noreply.github.com> Date: Thu, 26 Feb 2026 20:08:50 +0530 Subject: [PATCH 3/3] Update CodingAgent.md --- GSoC-Ideas/2026/CodingAgent.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/GSoC-Ideas/2026/CodingAgent.md b/GSoC-Ideas/2026/CodingAgent.md index 2010912..f638357 100644 --- a/GSoC-Ideas/2026/CodingAgent.md +++ b/GSoC-Ideas/2026/CodingAgent.md @@ -2,7 +2,7 @@ ## Background and Motivation Our motivation for building CodingAgent is that there is still no truly good open-source coding agent that works like the best closed-source tools such as Codex, Antigravity, or Claude Code. -We’ve used all of these extensively and we are frustrated that I can’t plug my favorite open-source models into a comparable, powerful agent experience. +We’ve used all of these extensively and we are frustrated that We can’t plug my favorite open-source models into a comparable, powerful agent experience. Key capabilities like robust persistent memory, solid MCP integration, and local vector-based storage for long-term context are either missing or poorly implemented in current open-source options. We want to change that by creating an agent that treats memory and context as first-class citizens, not afterthoughts. My goal is to build something that operates at the level of serious engineering workflows, not just toy demos.