zyker

Running in production · onboarding design partners

Ship with agents.
Keep control.

The project system for teams that put AI agents to work. Assign an issue to Claude Code or Codex, run it on your infrastructure, and watch every question, command, diff, test, cost, and handoff arrive on one accountable timeline.

Your repositories and LLM keys stay on your machines. zyker charges for orchestration — never marked-up inference.

In Progress

Align the holder chips

Claude Code · running on your VPS

To do

Backfill org memberships

Draft the pricing page

Done

Close the MCP write hole

landed · awaiting review
  1. Claude Code moved todo → doing 
  2. ✗ chips drifting off the column · traced to an in-flow pip
→ hoisted the pip into a gutter
✓ mix precommit — 1484 tests passed
  3. Claude Code +88 −15 · landed on main · awaiting review
  4. Opus 5 · $2.75 · 14 min

real run data from zyker's own board — ZYK-423

Apply for design partnership 3–5 teams · founder-led setup

One issue, fully accountable

01

Delegate the issue

Choose Claude Code or Codex, pick a model, and assign the work like you would to a teammate.

02

Run on your infrastructure

A signed task reaches your runner. The agent works in a locked container with your repository and keys.

03

Review the receipt

Progress, questions, logs, diffs, tests, cost, and CI status return to the issue as they happen. run_completed · +88 −15 · $2.75

Built for the team that already owns its infrastructure

Your control plane

zyker.dev

Board · briefs · documents · portfolio

Plan work where humans and agents share the same issues. See what is live across projects, answer an agent's blocking question, review its outcome, and keep the complete execution record beside the work.

The SaaS stores orchestration state and the event ledger — never your model credential.

Your execution plane

your runner

Your VPS · your repository · your keys

A small outbound-only runner receives a signed task and starts Claude Code or Codex in a locked container. You choose the model, toolchain, workspace, and egress policy. Your code and credentials stay under your control.

No inbound ports. No shared SSH keys. No inference resale.

Why the runner is yours

  • Every task payload is Ed25519-signed by the control plane and verified byte-for-byte by the runner before it's parsed.
  • The runner dials out over a single outbound WebSocket — no inbound ports, no shared SSH keys, works behind NAT.
  • Agent runs execute in a container on an egress-deny network behind an allowlist proxy.
  • Your LLM keys are read on your machine and never stored in zyker's database; run credentials are minted per dispatch, scoped to one issue.

Linear lets you assign issues to someone else's agents on someone else's cloud. zyker orchestrates your agents on your servers, on your keys.

zyker tracks zyker

We build in the open and dogfood it — this product's own backlog is filed and worked through its own agent API. The honest state of things:

Live today

  • Headless agent runs on your own VPS — executing zyker's backlog since July 18
  • Runner events streaming onto the issue timeline — logs, diffs, questions, cost
  • Per-run cost and model, from real usage on every run
  • Outcome receipts — every agent landing shows who reviewed it
  • The board, briefs, and the agent API (MCP) your Claude Code / Codex talks to

In development

  • One-command runner setup — provisioning still takes developer hands
  • Pull-request handoff polish — trunk handoff is solid; PR-open is being hardened

Join the waitlist

We're onboarding a small number of design partners. Tell us if you fit.

Design partners: we provision your first runner with you on a call, you watch an agent ship a real change on your own VPS in the first session, and you get the founder on a direct line. Free through the partnership — in return, honest feedback. Taking 3–5 teams.

Local list only — no mailing-list integration, no spam. One email when we're ready for you.