Skip to content

AgentOSOnYourCodebase.md

Latest commit

 

History

History
52 lines (35 loc) · 3.62 KB

AgentOSOnYourCodebase.md

File metadata and controls

52 lines (35 loc) · 3.62 KB

The Agent OS on Your Codebase

Today, in public, Neo's engineering team maintains the very engine it runs on. The same Agent OS is being built to inhabit your codebase — one the models were never trained on.

The hardest problem in applying AI to a real codebase is not generating code; it is understanding the code that already exists — its history, its conventions, its load-bearing decisions. A general model knows its training data. It does not know your repository. Neo's Brain is built to close exactly that gap.

What makes it portable

The capabilities that let the swarm maintain Neo are not Neo-specific:

  • The Knowledge Base semantically ingests a repository, so agents can ask grounded questions about your code instead of pattern-matching from training data.
  • The Memory Core & Native Edge Graph give agents persistent, structural understanding that accumulates across sessions — your architecture, your tickets, your decisions, as a queryable graph.
  • The cross-family engineering team and its reviewed lifecycle are codebase-agnostic: the process that catches blind spots on Neo catches them anywhere.

Ingestion is driven by an explicit tenant model — point the Agent OS at a repository and it builds its understanding from that source. (Mechanism: Tenant Ingestion Model.)

flowchart LR
    classDef src fill:#222,stroke:#f5a623,stroke-width:2px,color:#fff
    classDef brain fill:#0f3460,stroke:#16c79a,stroke-width:2px,color:#fff
    classDef team fill:#4a1942,stroke:#e74c3c,stroke-width:2px,color:#fff

    Repo["Your codebase (any stack)"]:::src
    KB["Knowledge Base + Memory Core ingest"]:::brain
    Understanding["Persistent structural understanding"]:::brain
    Team["Cross-family engineering team"]:::team

    Repo --> KB --> Understanding --> Team
    Team -.->|"reviewed changes"| Repo
Loading

What's proven vs. what's being built

This is a capability story with honest boundaries — not a present-tense promise:

  • Proven, in the open today: the Agent OS autonomously maintains Neo.mjs itself — memory, PRs, cross-family review, the Dream Pipeline, and self-healing workflows, under a human merge-gate.
  • Portable trajectory: the same Agent OS is being shaped to ingest, reason over, and help maintain other codebases, across languages.
  • Boundary: live-runtime possession through the Neural Link is proven for the Neo runtime; reaching into arbitrary language runtimes is a future extension, not a current claim.

The honest version of the pitch is the strong one: it maintains its own codebase today; it is being built to inhabit yours.

Why this is the differentiated half

A code-generation assistant gives you output. An Agent OS that learns your codebase gives you a team with memory — one that retains why a decision was made, reviews its own work across model families, and gets better the longer it operates on your system. That is the capability the underlying models cannot provide alone, and it is the half of Neo most worth deploying.

Go deeper