kiss-agent-framework 2026.6.11

KISS Agent Framework - A simple and portable agent framework for building and evolving AI agents

KISS Sorcar

The open-source AI coding agent that beats Cursor and Claude Code on Terminal Bench.

Terminal Bench 2.0:  KISS Sorcar 62.2%  •  Cursor agent 61.7%  •  Claude Code 58%

Free. Local. Bring your own API key. Runs as a VS Code extension and as a web/mobile app.

"Everything should be made as simple as possible, but not simpler." — Albert Einstein

Website: https://kisssorcar.github.io/ · Paper: arXiv:2604.23822

---

Table of Contents

• Introduction to KISS Sorcar
• See It in Action
• Full Installation
• KISS Sorcar Extension Installation
• CLI Interface
• Messaging & Third-Party Agents
• Models Supported
• Contributing
• License
• Citation

Introduction to KISS Sorcar

KISS Sorcar is an open-source AI software engineering and general-purpose assistant, which is free, open-source, and fully local. You bring your own model API key (Anthropic highly recommended) or LLM endpoint or Claude Code or OpenAI Codex; nothing about your code or prompts is sent through our servers.

Full Installation

curl -fsSL https://raw.githubusercontent.com/ksenxx/kiss_ai/main/scripts/install.sh | bash

KISS Sorcar runs as a Visual Studio Code extension, as a CLI tool and as a web/mobile app, and is built on the KISS Agent Framework — a deliberately simple agent runtime that follows the KISS principle ("Keep it Simple, Stupid"). All agents run as daemon thread in the KISS Sorcar web server daemon process. The agent has browser support (Chromium + Playwright), multimodal support, Docker container support, can research topics on the web, and can run for hours across multiple sessions. If you have Claude Code or OpenAI Codex in your PATH, you can also use cc/* or codex/* models for chat.

KISS Sorcar is named after P. C. Sorcar, the Bengali magician. The paper is at papers/kisssorcar/kiss_sorcar.pdf.

Engineering principles are encoded in the agent's system prompt. See Section 5 of the paper for details.

An old video on KISS Sorcar can be found at https://www.youtube.com/watch?v=xnYxWvRqACE. We no longer recommend explicitly creating a plan in KISS Sorcar. See the paper for details.

Note that Sorcar also means government in Bengali.

🎬 See It in Action

From writing production-grade code to planning your next vacation, KISS Sorcar handles a range of tasks end-to-end. Here are a few examples:

💻 Coding & Software Engineering Writes, debugs, and refactors code.	✈️ Trip Planning & Research Browses the web, compares options, and assembles itineraries.
💬 Desktop & Messaging Apps Drives native apps like Slack via the desktop, end-to-end.	📱 Mobile/Web App You can use KISS Sorcar as a mobile/web app.

KISS Sorcar Extension Installation

To install the KISS Sorcar extension only, open Visual Studio Code, search for "KISS Sorcar" in the extension marketplace, install, and relaunch VS Code. Press ESC if you don't have a specific API key, but you must provide at least one API key.

You can also manually download the extension from src/kiss/agents/vscode/kiss-sorcar.vsix.

CLI Interface

If you do not want to use the KISS Sorcar IDE, you can open a terminal and use sorcar as a shell command or a CLI (similar to Claude Code). Some examples:

sorcar # to launch Sorcar CLI similar to Claude Code

sorcar -t "What is 2435*234"

sorcar -n -t --use-chat "What is 2435*234?" # start a new chat session

sorcar -m "claude-sonnet-4-6" -t "What is 2435*234?" # use a specific model

echo "Can you find the cheapest non-stop flight from SFO to JFK on June 15?" > prompt
sorcar -f prompt # use contents of a file as the task

sorcar -t 'Can you send the message "Hello from Sorcar!" to ksen via the desktop slack app?'

sorcar -t 'Can you show me the detailed step-by-step workflow of gepa.py?'

CLI Options

Flag	Description
-t, --task	Task description
-f, --file	Path to a file whose contents to use as the task
-m, --model_name	LLM model name (default: claude-opus-4-6)
-e, --endpoint	Custom endpoint for a local model
--header	Custom HTTP header in Key:Value form; may be repeated
-b, --max_budget	Maximum budget in USD
-w, --work_dir	Working directory
-v, --verbose	Print output to console (default: true)
-n, --new	Start a new chat session
-c, --chat-id	Resume a chat session by ID
-l, --list-chat-id	List the last 10 chat sessions with tasks and results
-p, --parallel	Enable parallel subagents
--no-web	Disable browser/web tools (terminal-only mode)
--use-chat	Use chat mode
--use-worktree	Use chat mode with git worktree isolation (advanced)
--cleanup	Scan for and clean up orphaned worktree branches

💬 Messaging & Third-Party Agents

KISS Sorcar includes 23 third-party messaging agents that can send and receive messages on your behalf:

BlueBubbles · Discord · Feishu · Gmail · Google Chat · iMessage · IRC · LINE · Matrix · Mattermost · Microsoft Teams · Nextcloud Talk · Nostr · Phone Control · Signal · Slack · SMS · Synology Chat · Telegram · Tlon · Twitch · WhatsApp · Zalo

It also ships a Govee smart-home CLI for controlling IoT lights (on/off, brightness, color, color temperature) via the Govee Developer API.

These agents are in src/kiss/agents/third_party_agents/.

🤖 Models Supported

529 models across 8 provider categories (OpenAI, Anthropic, Gemini, Together AI, MiniMax, OpenRouter, Claude Code CLI, Codex CLI) with built-in pricing, context lengths, and capability flags.

Generation Models (text generation with function calling support):

• OpenAI: gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o, gpt-4o-mini, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, gpt-5.1, gpt-5.2, gpt-5.2-pro, gpt-5.3-chat-latest, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-pro, gpt-5.5, gpt-5.5-pro (+ dated variants)
• OpenAI (Codex): gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini, gpt-5.2-codex, gpt-5.3-codex
• OpenAI (Reasoning): o1, o1-pro, o3, o3-mini, o3-pro, o3-deep-research, o4-mini, o4-mini-deep-research (+ dated variants)
• OpenAI (Image): gpt-image-1, gpt-image-1-mini, gpt-image-1.5, gpt-image-2
• OpenAI (Search): gpt-4o-search-preview, gpt-4o-mini-search-preview
• OpenAI (Other): computer-use-preview, openai/gpt-oss-20b, openai/gpt-oss-120b
• Anthropic: claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5, claude-opus-4-1, claude-opus-4, claude-sonnet-4-6, claude-sonnet-4-5, claude-sonnet-4, claude-haiku-4-5 (+ dated variants)
• Claude Code CLI: cc/haiku, cc/opus, cc/sonnet
• Codex CLI: codex/default, codex/codex-auto-review, codex/gpt-5.2, codex/gpt-5.3-codex, codex/gpt-5.4, codex/gpt-5.4-mini, codex/gpt-5.5
• Gemini: gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-image, gemini-2.5-flash-lite, gemini-2.0-flash, gemini-2.0-flash-lite, gemini-3.5-flash, gemini-3.1-flash-lite
• Gemini (Preview): gemini-3-pro-preview, gemini-3-flash-preview, gemini-3.1-pro-preview, gemini-3.1-flash-lite-preview, gemini-3.1-flash-tts-preview
• Gemini (Image): gemini-3-pro-image, gemini-3.1-flash-image
• Gemini (Open Models): google/gemma-4-31B-it, google/gemma-3n-E4B-it, google/gemma-2-27b-it
• Together AI (Llama): Llama-4-Scout, Llama-4-Maverick (with function calling), Llama-3.x series (generation only)
• Together AI (Qwen): Qwen2-1.5B-Instruct, Qwen2-VL-72B-Instruct, Qwen2.5-72B/14B/7B-Instruct, Qwen2.5-VL-72B, Qwen2.5-Coder-32B, Qwen3-235B series, Qwen3-Coder-480B, Qwen3-Coder-Next, Qwen3-Next-80B, Qwen3-VL-32B/8B, Qwen3.5-397B/9B, Qwen3.6-Plus, QwQ-32B
• Together AI (DeepSeek): DeepSeek-R1, DeepSeek-R1-0528, DeepSeek-R1-0528-tput, DeepSeek-R1-Distill-Llama-70B, DeepSeek-R1-Distill-Qwen-1.5B/14B, DeepSeek-V3-0324, DeepSeek-V3.1, DeepSeek-V4-Pro, deepseek-coder-33b-instruct
• Together AI (Kimi/Moonshot): Kimi-K2-Instruct, Kimi-K2-Instruct-0905, Kimi-K2-Thinking, Kimi-K2.5, Kimi-K2.6
• Together AI (Mistral): Ministral-3-14B, Mistral-7B-v0.1/v0.2/v0.3, Mistral-Small-24B, Mixtral-8x7B
• Together AI (Z.AI): GLM-5, GLM-5.1, GLM-4.5-Air-FP8, GLM-4.6, GLM-4.7
• Together AI (MiniMax): MiniMax-M2.5, MiniMax-M2.7
• MiniMax: minimax-m2.5, minimax-m2.5-lightning (via MiniMax direct API)
• Together AI (DeepCogito): cogito-v1-preview (llama-70B/8B/70B-Turbo, qwen-14B/32B), cogito-v2-1-671b
• Together AI (NVIDIA): Llama-3.1-Nemotron-70B, Nemotron-Nano-9B-v2
• Together AI (Other): arcee-ai/trinity-mini, essentialai/rnj-1-instruct
• OpenRouter: Access to 313 models from 53 providers via unified API, including OpenAI, Anthropic, Google, Meta Llama, DeepSeek, Qwen, Amazon Nova, Cohere, X.AI Grok, MiniMax, ByteDance/ByteDance Seed, MoonshotAI, Mistral, NVIDIA, Z.AI/GLM, AllenAI, Perplexity, NousResearch, Baidu ERNIE, Xiaomi, Reka AI, Arcee AI, Perceptron, StepFun, Tencent, Inception, InclusionAI, KwaiPilot, Liquid, Morph, Relace, IBM Granite, Prime Intellect, AI21, Aion Labs, Anthracite, DeepCogito, Essential AI, Gryphe, Inflection, Mancer, Microsoft, Sao10K, Switchpoint, TheDrummer, Undi95, Upstage, Writer, and others

Embedding Models (for RAG and semantic search):

• OpenAI: text-embedding-3-small, text-embedding-3-large, text-embedding-ada-002
• Google: gemini-embedding-001, gemini-embedding-2, gemini-embedding-2-preview
• Together AI: BAAI/bge-base-en-v1.5, intfloat/multilingual-e5-large-instruct

Each model entry includes capability flags: is_function_calling_supported, is_generation_supported, is_embedding_supported.

🤗 Contributing

Contributions in the form of issues are welcome! KISS Sorcar should be able to take care of them.

📄 License

Apache-2.0

📚 Citation

If you use KISS Sorcar in your research, please cite:

@misc{sen2026kisssorcar,
  title         = {KISS Sorcar: A Stupidly-Simple General-Purpose and Software Engineering AI Assistant},
  author        = {Sen, Koushik},
  year          = {2026},
  eprint        = {2604.23822},
  archivePrefix = {arXiv},
  primaryClass  = {cs.SE},
  url           = {https://arxiv.org/abs/2604.23822}
}