swink: to toil, to labor, to work hard
Swink Agent
An extensible policy-minded performant agent harness, in pure Rust.
Any LLM, one trait
Implement StreamFn to plug in Anthropic, OpenAI, Gemini, Ollama, or your own.
Learn more →
Tools run concurrently
Batch tool calls execute in parallel. Steering callbacks can interrupt mid-batch.
Learn more →
No unsafe. No globals.
#[forbid(unsafe_code)] at every crate root. Pure, predictable state.
Learn more →
Policy guardrails, not afterthoughts
Budget caps, sandboxing, PII redaction, prompt injection detection — compiled into the loop, not bolted on. Zero latency. No extra LLM calls.
Learn more →
Batteries included
13 crates in one workspace: 9 LLM adapters, MCP integration, session memory, eval harness, artifact storage, orchestration patterns, on-device inference, proc macros, and a full TUI.
Learn more →
swink-agent
swink-agent-adapters
swink-agent-policies
swink-agent-memory
swink-agent-local-llm
swink-agent-eval
swink-agent-artifacts
swink-agent-auth
swink-agent-mcp
swink-agent-patterns
swink-agent-plugin-web
swink-agent-macros
swink-agent-tui
Swink Agent Examples
Runnable examples covering common patterns: a simple prompt loop, tool registration with approval callbacks, implementing a custom
StreamFn adapter, and a full agent wired to the TUI.
View examples on GitHub →