Developer toolkit for Context Runtime

The agent toolkit the ReDevOps reference applications build on — LLM client, tool registry, approval flow, sandboxed execution, guardrails and an eval harness. The reference applications pair it with Context Runtime, which decides what context each agent sees before it runs.

Context Runtime  →  ReDevOps RAG  →  Sidekick  →  Application logic

---

agent_harness

🚀 NVIDIA Inception Program Member — ReDevOps is a member of the NVIDIA Inception Program, supporting startups advancing AI and accelerated computing. Membership provides access to NVIDIA technology, technical resources, and the startup ecosystem. It does not imply product endorsement by NVIDIA.

agent_harness is the OSS core that provides the building blocks for a safe, tool-using agent: LLM client, tool registry, approval flow, sandboxed execution, guardrails, and an eval harness. It is the foundation that an agent layer consumes.

Quickstart

pip install -e .
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_API_KEY=ollama
export MODEL=llama3.1
python -m examples.tiny_agent

Note: the bundled tiny_agent example runs fully offline against the stub LLM client (it echoes the prompt rather than calling a model), so the env vars above only take effect once an agent layer wires a real provider.

Using the modules

from agent_harness import llm, tools, approval, sandbox, guardrails

Architecture

agent_harness (OSS core) + thin agent layer that wires real providers: llama.cpp/ollama, vLLM, Moonshot/Kimi, xAI/Grok, OpenAI. See docs/architecture.md and docs/configuration.md.

License

AGPL-3.0