The Weev agent

One outbound-only agent, installed as a service on your own servers. Within the hour, on a read-only change ticket, it has discovered the jobs your schedulers already run and is streaming their runs into Weev as evidence. Then each job escalates at its own pace.

Design preview · Upcoming

The agent is a design, not a shipped feature — and the design itself is pending customer discovery, so details here are more likely to change than Link Mode's. Nothing on this page is installable. The shipped path today is the SDK Mode quickstart.

The shape

Install once, outbound only

A single binary running as a Windows service. It dials out over 443 to one Weev hostname — no listener, no inbound port, no VPN, proxy-aware. It runs unprivileged, needing only the permissions the jobs it observes already need.

Enrollment (design)Previewshell

$ weev-agent enroll --token wv_7f2a
  service installed · outbound 443 only
  discovered 41 jobs (sql-agent 28, tasks 13)
  streaming runs · hash chain intact

The rungs

Observe. Adopt. Own. Native.

The agent's role is chosen per job and recorded per run, and every step has a mechanical rollback. Observation is the first shipped mode in the design; most jobs never need the top rung.

Observe

Read-only. Discovers jobs in SQL Server Agent and Task Scheduler and streams every run into Weev as a tamper-evident record.

Adopt

Wraps a job. The incumbent stays the execution engine; Weev owns the clock, not the run. Rollback is one command.

Own

Weev fires the job through the agent on its declared schedule and dependencies. The incumbent's entry retires.

Native

Full lifecycle — held, leased, heartbeated and accounted for like any SDK job.

The rules

What it will never do

The agent never executes a command string received from the cloud — only job identity and an attempt token cross the wire inbound. Observations are written to a hash-chained store on disk before transmission; credentials never leave the box; and the design ships as open source so those rules are verifiable rather than asserted. The full security story lives on the Security & Architecture page.