Back to Home
Project 04

AI Agents

Composable task-agent framework

A framework for AI agents that execute real tasks via structured tools, guardrails, and an audit trail. Separates proposal from execution and handles retries safely.

Role
Systems designer
Status
Open source
Period
2024 — Now
Domain
AI infrastructure
LLM OrchestrationTool-callingPydantic v2StreamingIdempotency
The Problem

Most agent frameworks blur the line between proposal and execution — agents either do too little (chat-only) or too much (silent side effects). What's missing is a disciplined runtime where every action is reviewable.

Approach

Separate reasoning from execution. Tools declare their effects ahead of time; the runtime stages a pending action that a human or policy approves before it runs. Retries are idempotent by construction.

Architecture

Tool contract

Pydantic v2 schemas describe inputs, side effects, and rollback semantics for every tool.

Runtime

Streaming LLM orchestration with deterministic tool dispatch, structured logs, and a replayable audit trail.

Guardrails

Policy hooks veto unsafe actions before they reach the proposer; idempotency keys make retries safe.

Extensibility

30+ tools shipped; a new tool lands via a single class plus a JSON spec — no runtime rewiring.

Outcomes
  • Zero unreviewed executions across thousands of agent steps.
  • Tool authors ship in minutes, not hours.
  • The repo is the proof — open source, ready to read.
Stack
LLM OrchestrationTool-callingPydantic v2StreamingIdempotency keysAudit trail
Screenshots
Runtime flow — propose, guardrail check, approve, then execute.