docsclaw · v0.3.0

Lightweight Agentic Runtime for OpenShift

Deploy AI agents with a system prompt, agent card, and tool config — no code changes needed. A2A-compatible, multi-provider, ~10 MiB per pod.

$ brew install pavelanni/tap/docsclaw

GitHub Get Started Demos Presentations

A2A Protocol OpenShift Multi-Provider LLM Tool Use Session Management

Why DocsClaw?

~10 MiB

per pod

A2A

protocol native

Zero

code changes

Lightweight Go Binary

Single static binary, ~10 MiB per pod on OpenShift. No Python or TypeScript runtime, no heavy dependencies. Starts in milliseconds.

A2A Protocol Native

Built on the Agent-to-Agent protocol. Agents discover, delegate, and orchestrate work with any A2A-compatible agent.

ConfigMap Personality

Agent behavior defined by a config directory: system prompt, agent card, tool config. Change personality by swapping a ConfigMap — no rebuild needed.

OCI Skill Images

Mount skills securely into the agent container from OCI images via skillimage.dev. Sigstore-signed, versioned, and delivered as Kubernetes image volumes.

Multi-Provider LLM

Anthropic, OpenAI, and any OpenAI-compatible API (LiteLLM, vLLM, Ollama). Switch providers with a config change. Token usage tracking built in.

Tool-Use Agentic Loop

Parallel tool execution via goroutines. Shell commands, file operations, web fetch, document service. Extensible tool registry with before/after hooks.

Server-Side Sessions

Multi-turn conversations with in-memory session management. The server maintains conversation history — no client-side state needed.

Quick Start

        # Build from source

        $ make build

        # Start the agent server

        $ LLM_API_KEY=sk-... bin/docsclaw serve \

          --config-dir testdata/standalone \

          --listen-plain-http

        INFO Listening addr=0.0.0.0:8000

        # Chat with the agent

        $ bin/docsclaw chat --agent-url http://localhost:8000/a2a

        # Or use the A2A CLI

        $ a2a discover http://localhost:8000

        $ a2a send http://localhost:8000 "What is a transistor?"

How It Works

DocsClaw is a universal agent harness. You provide the personality via configuration files; it provides the runtime, protocol handling, and tool execution.

You Provide

system-prompt.txt — agent personality
agent-card.json — A2A discovery metadata
agent-config.yaml — tools and loop settings

DocsClaw Provides

A2A protocol server
LLM provider abstraction
Agentic tool-use loop
Session management
Metrics and logging