Learn and solve problems with AI tools.

Open AI tooling for real infrastructure: agent pipelines, MCP surfaces, and GitOps platforms, aligned with software freedom.

DataKnifeAI logo: a circuit-edged data knife inside an olive hexagon

North star

A fully autonomous agent pipeline that can hand off work to run in the background, using API LLMs and local LLMs, so people set direction while agents execute, monitor, and iterate without constant babysitting.

Where the north star is already showing up in code.

Nauarchos concept art: an admiral directing a fleet of agent vessels

Agent pipeline

Issues in, agents out

An issue opened on GitHub or GitLab is triaged by a small model, escalated when confidence is low, and, if it is real work, handed to a coding agent inside a fresh Coder workspace. The result comes back as a comment on the issue.

  1. WebhookGitHub / GitLab issue
  2. Triagegpt-5-nano, escalate below 0.7
  3. BerthFresh Coder workspace
  4. DispatchCursor CLI coding agent
  5. ReportComment and label
  • slashbay Active

    Intake, triage, berth, and dispatch.

  • dioptra

    Observe-only fleet status board.

  • nauarchos Design

    Rust control plane for agent fleets on Kubernetes.

Infrastructure monitor · Rust · MCP

pulseblade Active

Open observability with human gates: node collectors, a hub fleet view, a Postgres change journal, and MCP so agents ask what moved instead of drowning in alert noise. Humans get the same state without an LLM (dashboard, API, Prometheus, pulseblade ctl). Not a product pitch; self-hosted tooling you run.

  • pulseblade Active

    Collectors, hub, journal, gated remediation. Apache-2.0.

  • proxmox-ve-mcp

    MCP surface for Proxmox VE alongside host and cluster collectors.

  • Storage note

    Platform work also includes TrueNAS CSI upstream engagement; forks are not the hero here.

Model serving · GPU · Kubernetes

Local and API LLMs on real hardware

Self-hosted inference on machines you own: vLLM for Hermes Agent on a local GPU, plus Kubernetes-ready GPU workloads when the cluster is the right shape. Cloud APIs stay available when reach matters more than locality.

  • enodios Active

    One-script Linux setup for Hermes Agent + vLLM self-hosting.

  • freya

    GPU ComfyUI, SwarmUI, and Ollama on Docker/Kubernetes.

  • agent-skills · agent-workspace

    Shared skills and Cursor Cloud Agent workspace pools.

Kubernetes operator · Go

palworld-operator Beta

One custom resource becomes a running Palworld dedicated server: Deployment, saves, services, and Envoy Gateway exposure, using the official Pocketpair image.

apiVersion: palworld.dataknife.ai/v1alpha1
kind: PalworldServer
metadata:
  name: world-keeper
spec:
  generateSecrets: true
  gateway:
    address: 192.0.2.40

API · MCP · UI

High Command Live

Real-time Helldivers 2 war data behind a FastAPI scraper, exposed to agents through an MCP server, and queried in plain language from a chat UI.

> War Status

high-command-ui → high-command-mcp → high-command-api

active campaigns, liberation progress, global statistics

> List Planets

> Show Factions

> Get Biomes

D.O.N.T. Ship poster: Wally the stork, Dock Supervisor at Box 407-B, holding a parcel marked Outcome Box 407-B hub: three doors with J.E.V. odds of 78, 18, and 4 percent Isometric courier run through a low-poly suburban pocket universe

Browser game · Three.js · No build step

D.O.N.T. Ship Live

Definitely Not On Time. Delivering Outcomes, Not Timelines. You are a courier for a cheerful multiverse parcel company. At Box 407-B the J.E.V., an AI, posts odds on three doors; you pick one, then earn it by crossing an isometric pocket universe to hand the parcel to the customer. AI sets the odds, a human makes the call.

  1. OddsJ.E.V. prices three doors
  2. PickYour call, your stake
  3. RunIsometric pocket universe
  4. DeliverParcel to the customer
  5. ScoreC.O.D. and O.D.D.S.
  • C.O.D.

    Your bankroll. It says you were lucky.

  • O.D.D.S.

    Your calibration. It says you were right.

  • dont-ship

    Vanilla JavaScript and Three.js, low-poly isometric, on GitHub Pages. Apache-2.0.

What we're building toward

Four layers that let agents do real work on real systems.

Local and API models

Self-hosted inference when you want agency, cloud APIs when you want reach.

enodios freya

In motion

Most recently pushed public repositories.

All projects

Loading recent activity…

Principles

Maintainable
Clear structure people can operate and evolve over time.
Automated
Repeatable workflows over one-off heroics.
Stable
Boring reliability over fragile cleverness.
Freedom-respecting
Open and free software that preserves user agency, when it fits.