jonwiggins/optio

Workflow orchestration for AI coding agent swarms, from task to merged PR.

What it solves

Optio is a self-hosted AI engineering platform designed for organizations that cannot ship their source code, secrets, or production data to third-party SaaS AI agents. It automates the lifecycle of software engineering tasks—from ticket intake to merged pull requests—while keeping all execution and logs within the the user's own Kubernetes cluster.

How it works

Optio operates as an orchestration layer on Kubernetes, managing three distinct tiers of AI work:

  • Tasks: Converts tickets (from GitHub, Jira, Linear, or Notion) into merged PRs. It provisions isolated pods, runs an agent to write code, monitors CI, and automatically resumes the agent to fix failures or address reviewer feedback until the PR is merged.
  • Jobs: Parameterized, one-shot agent runs in isolated pods without a repository checkout, used for reports, triage, or scheduled automation.
  • Agents: Long-lived, message-driven processes that can communicate with each other via an internal API to form multi-agent teams.

The system uses a Kubernetes-style reconciliation control plane to ensure tasks don't get stuck and supports multiple agent vendors (Claude, OpenAI, Gemini, etc.) via adapters. It also uses "Connections" to inject external tools (like Slack, PostgreSQL, or MCP servers) into agent pods at runtime.

Who it’s for

  • Security-conscious organizations and regulated industries (finance, healthcare, government) that require strict data residency and tenancy isolation.
  • Teams already using Kubernetes who want a drop-in AI orchestration layer.
  • Multi-agent shops who want to avoid vendor lock-in by switching between different LLM providers per repository.
  • Platform teams building internal AI tooling.

Highlights

  • Autonomous feedback loop: Automatically handles CI failures and code review requests by resuming the agent with the necessary context.
  • Self-hosted infrastructure: Runs entirely in the user's K8s cluster, ensuring code and secrets never leave the network.
  • Multi-vendor support: Compatible with Claude Code, OpenAI Codex, GitHub Copilot, Google Gemini, and OpenCode.
  • Reconciliation control plane: A robust decision-loop architecture that prevents state loss and ensures task completion.
  • Inter-agent messaging: Enables persistent agents to collaborate as a team.
  • Integrated ticket intake: Native synchronization with GitHub Issues, Linear, Jira, and Notion.