InnerOS Engineering Journal

InnerOS role: Public Evidence / Engineering Notes
Lifecycle: Ongoing
Lineage: Public record of InnerOS architectural decisions, failures, measurements, and lessons learned.

The journal exists to make the engineering process inspectable, including failed approaches and the evidence that changed the architecture.

InnerOS principle: hackathons are validation environments. Reusable capabilities are extracted into maintained products and platform layers rather than treated as disconnected one-off projects.

Building a local-first, on-site AI software factory in public.

This repository documents the engineering journey behind InnerOS Local Agent Fabric: multi-agent coordination, local models, A2A, MCP, durable tasks, watchdogs, resource routing, browser automation, infrastructure operations and measured productivity.

The goal is not to publish a polished fairy tale. The failures matter. Wrong routing, false PASS states, missing handoffs, stale locks, source-of-truth drift and model limitations are recorded because each one changed the architecture.

Principles

• Local-first execution whenever practical.
• External models are escalation paths, not the default runtime.
• A2A for agent-to-agent interoperability.
• MCP for agent-to-tool and agent-to-resource access.
• Durable task state, locks, evidence and rollback.
• No PASS without real writes, real checks and traceable evidence.
• Measure human time saved separately from agent runtime.
• Public notes are sanitized. Credentials, client data and sensitive infrastructure details stay private.

Current series

1. Why InnerOS is adding A2A after building its own agent coordination layer
2. Failures that improved the Local Agent Fabric
3. Local models vs external coding agents: where model quality still matters
4. MCP and A2A are complementary, not competitors
5. Measuring the real ROI of agentic automation

Architecture direction

                     InnerOS
                        |
                 Resource Fabric
                        |
                       A2A
             +----------+----------+
             |          |          |
        Coding Agent  Repair    Guardian
             |          |          |
            MCP        MCP        MCP
             |          |          |
          Git/FS     Tools/APIs   Tests

The A2A protocol is an open standard for interoperability between agents, originally developed by Google and now hosted by the Linux Foundation. The project reached a stable production-ready v1.0 in 2026.

Official references:

• https://a2a-protocol.org/
• https://a2a-protocol.org/v1.0.0/
• https://www.linuxfoundation.org/press/linux-foundation-launches-the-agent2agent-protocol-project-to-enable-secure-intelligent-communication-between-ai-agents

Status

Active engineering journal. Entries are added from real development evidence and sanitized before publication.