v0.2.0 — Now with Media RAG + Rollback

Autonomous AI agents.
No cloud. No keys. No limits.

SumoSpace is a locally-first agent framework with a three-agent safety committee, built-in rollback, and zero cloud dependencies.

Get started → View on GitHub

pip install sumospace

🛡️

Three-Agent Safety Committee

Planner, Critic, Resolver deliberate before execution

🏠

Truly Local

HuggingFace, Ollama, vLLM. No cloud required.

↩️

Rollback & Snapshots

One command to undo any agent action

🎯

Media RAG

Ingest images, audio, video alongside code

🔌

Plugin Ecosystem

Register tools via pip install

📊

Built-in Benchmarks

Reproducible evaluation framework

flowchart LR
    User([User]) --> Kernel[Kernel]

    subgraph Committee ["Safety Committee"]
        direction TB
        Planner[Planner] --> Critic[Critic]
        Critic -- Reject --> Resolver[Resolver]
        Resolver -. Revise .-> Planner
        Critic -- Approve --> Approved((Plan))
    end

    Kernel --> Classifier{Classifier}
    Classifier --> Committee

    Approved --> Executor[Tool Executor]
    Executor --> Snapshot[(Snapshot)]
    Snapshot --> Memory[(Memory)]
    Memory --> Kernel

    Kernel --> Output([Final Answer])

Ollama (Local)CLIOpenAI

from sumospace import SumoKernel, SumoSettings
import asyncio

async def main():
    async with SumoKernel(SumoSettings(
        provider="ollama",
        model="phi3:mini",
    )) as kernel:
        trace = await kernel.run(
            "Add docstrings to all functions in ./src/utils.py"
        )
        print(trace.final_answer)

asyncio.run(main())

pip install sumospace
ollama pull phi3:mini
sumo run "Add docstrings to all functions in ./src/utils.py"

async with SumoKernel(SumoSettings(
    provider="openai",
    model="gpt-4o",
    # reads OPENAI_API_KEY from env
)) as kernel:
    trace = await kernel.run("Refactor auth.py to use async/await")

Why SumoSpace?¶

Feature	SumoSpace	LangChain / LlamaIndex
Core Philosophy	Execution-first, safe autonomy	Chain-building, orchestration
Built-in Safety	✔ Three-Agent Committee	✘ Build your own
Rollback Support	✔ Native filesystem snapshots	✘ None
Cloud Independence	✔ 100% offline capable	~ Possible, but cloud-native
Multi-Modal RAG	✔ Native audio/video/image	✔ Supported

Benchmarks

We run reproducible benchmarks using our built-in framework. Results for phi3:mini and llama3:8b will be published here after community validation. Run your own:

sumo benchmark compare --provider ollama --model llama3:8b

Autonomous AI agents. No cloud. No keys. No limits.

Why SumoSpace?¶

Ready to run your first agent?

Autonomous AI agents.
No cloud. No keys. No limits.