Releases·Tool Release
pi-loop-studio/orchestra: source-level architecture of an evidence-driven, recoverable multi-agent runtime built on XState and SQLite
The pi-loop-studio/orchestra project published a source-level technical architecture report on GitHub, describing a ground-up rebuild of its predecessor pi-agent-loop. The system is an evidence-driven, recoverable multi-agent engineering runtime that makes the scout-plan-execute-review-verify-publish pipeline an explicit, persistent, auditable state machine. It matters to practitioners building parallel agent workflows because it emphasizes recoverability, isolation, and mechanical verification rather than UI features.
39 engagement1 sourceFri, Sep 11, 2026, 09:47 AM
AI Overview
- pi-loop-studio/orchestra rebuilds pi-agent-loop as an explicit XState workflow: scout, plan, parallel execute, independent review, verify, repair, single-writer publish.
- PiRuntime assigns fixed global slots: 1 Principal chair, 20 Executor workers, 8 Auxiliary/review roles, on @earendil-works/pi-agent-core.
- Persistence uses SQLite WAL as a transactional fact store with isolated git worktrees and single-writer publication.
- UI is React 19 + Vite with @assistant-ui chat, React Flow dependency graphs, served by an Express 5 localhost HTTP/SSE API.
Related
Releases·Agent Release
IvyClaw: multi-agent AI coding system built on DeepAgents/LangGraph
The IvyClaw project published a production-oriented multi-agent system for software engineering tasks, built on DeepAgents and LangGraph. It orchestrates Planner, Researcher, Coder, Tester, and Reviewer agents through a full plan-retrieve-code-test-review workflow. For practitioners, it offers a reference architecture for running coding agents in production, with sandboxed execution, state persistence, async task handling, and observability built in.
The IvyClaw project published a production-oriented multi-agent system for software engineering tasks, built on DeepAgents and LangGraph. It orchestrates Planner, Researcher, Coder, Tester, and Reviewer agents through a full plan-retrieve-code-test-review workflow. For practitioners, it offers a reference architecture for running coding agents in production, with sandboxed execution, state persistence, async task handling, and observability built in.
56 engagement1 sourcegithubSun, Sep 13, 2026, 03:14 AM
Products·building
Event-driven AI agent with no while loop
Pizza is a personal AI agent that uses an event-driven architecture instead of a traditional while(true) loop. It logs every action as an event in a local SQLite store, and the agent's behavior is driven by an event-handler table, enabling interrupts, retries, and parallel tool calls. This design aims to make the agent more robust and maintainable.
Pizza is a personal AI agent that uses an event-driven architecture instead of a traditional while(true) loop. It logs every action as an event in a local SQLite store, and the agent's behavior is driven by an event-handler table, enabling interrupts, retries, and parallel tool calls. This design aims to make the agent more robust and maintainable.
9 engagement1 sourceredditMon, Aug 17, 2026, 02:30 PM
Releases·Tool Release
AI Graph Studio launches as free, open-source browser tool for designing and simulating agentic AI workflows
AI Graph Studio is a free, open-source, browser-based visual architecture tool for designing, documenting, analyzing, simulating, and optimizing agentic AI workflows and graphs. It targets the shift from prompt-based interactions toward structured AI systems composed of agents, tools, and routers. The project is hosted on GitHub, giving practitioners a visual canvas for agentic system design.
AI Graph Studio is a free, open-source, browser-based visual architecture tool for designing, documenting, analyzing, simulating, and optimizing agentic AI workflows and graphs. It targets the shift from prompt-based interactions toward structured AI systems composed of agents, tools, and routers. The project is hosted on GitHub, giving practitioners a visual canvas for agentic system design.
26 engagement1 sourcegithubMon, Sep 7, 2026, 08:47 AM
Products·building
Software factory with 6 scoped agents and orchestrator
A reference implementation of an end-to-end software development workflow using multiple AI agents. It takes work from GitHub, Sentry, deployment events, and manual requests, then routes it through triage, implementation, validation, release, documentation, and monitoring. Built in TypeScript with Mastra, it aims to automate and coordinate software delivery.
A reference implementation of an end-to-end software development workflow using multiple AI agents. It takes work from GitHub, Sentry, deployment events, and manual requests, then routes it through triage, implementation, validation, release, documentation, and monitoring. Built in TypeScript with Mastra, it aims to automate and coordinate software delivery.
15 engagement1 sourceredditWed, Aug 26, 2026, 08:33 PM
Social·Community
Karpathy advocates graph-based agent architecture over sequential loops
Andrej Karpathy's perspective on agent architecture, shared via X, argues that moving from a single sequential execution loop to a connected graph topology improves agent context and task orchestration. The post claims this shift made his loop 1000x better, resonating with engineers who see linear pipelines hit limits in complex state management.
Andrej Karpathy's perspective on agent architecture, shared via X, argues that moving from a single sequential execution loop to a connected graph topology improves agent context and task orchestration. The post claims this shift made his loop 1000x better, resonating with engineers who see linear pipelines hit limits in complex state management.
71 engagement2 sourcesredditWed, Aug 5, 2026, 03:25 AM