What happened
WakeVane’s website describes the product as an open-source harness for keeping coding agents working on a repository when a developer is away. It is built on Databricks’ Omnigent and is currently alpha. WakeVane calls itself a distribution of Omnigent, adding event subscriptions, a shim registry and an efficiency scorecard to a shared server for coding agents. It keeps the upstream omnigent CLI name to support compatibility and ongoing merges.
The event model is intended to start work when something happens rather than through a polling loop. The site gives a failed CI check, a new issue and an incoming webhook as triggers. In its software-factory example, a small model triages a failed check, then routes work to a fixer in a fresh sandbox. That agent returns a proposed change with tests and a recording; a human reviews the evidence before merging. After a merge, a watcher can respond to a deploy event, run health checks and report a verdict. WakeVane says tasks can continue in cloud sandboxes and sessions can be resumed from a terminal, browser or phone.
The website lists Claude Code, Codex, Cursor, OpenCode, Pi and custom agents as choices. It attributes the multi-agent runtime, cloud sandbox support across 12 providers, sessions spanning devices, policies, approvals and spend controls to Omnigent. WakeVane’s additions include subscriptions to events and a registry for “shims”: patches over model weaknesses, each with a reason, a review-by release and a record of when it fires. A janitor agent can open pull requests to remove shims that are no longer useful.
The scorecard exposes cache hit rate, cost per session, model mix, expensive sessions and a harness-and-model breakdown. The company presents these as ways to inspect and manage usage, not as guaranteed savings. Its website advises a focused pilot: try the factory on CI-failure events for a week, review fixes and inspect the scorecard. The displayed sequence is explicitly illustrative, not live activity. WakeVane says its first published run on a public repository—showing the triggering check, diagnosis, diff, rerun, elapsed time, session cost and remaining reviewer decisions—is not yet available. The site gives no price in the described material.
What the archive adds
New Runtime’s earlier coverage has described ECC packaging an engineering harness for coding agents as a reusable process rather than a prompt recipe (October 21, 2026). WakeVane is a new release, not a restatement: it makes event-triggered execution, cloud runtime and spending visibility part of an Omnigent-based distribution.
Trend signal
This supports harness architecture outlives model choice: WakeVane presents triggers, sandboxes, permissions, evidence and cost visibility as the operating layer around a choice of coding agent. The alpha site describes the architecture, but its first public real-run evidence is still pending, so the operational value is not yet independently demonstrated.
Why now
The product is aimed at work that otherwise waits for a person to notice a failed check, issue or deployment event. Its suggested one-week CI pilot makes the intended timing concrete: start investigations when signals arrive, then judge the workflow on returned evidence and actual session cost.
Why it matters
For a team adopting coding agents, unattended execution changes the operating question from “Can the model suggest a fix?” to “What can it do while nobody is watching, and what must a reviewer inspect before accepting the result?” WakeVane’s policies, ask-before-action approvals and spend caps offer explicit controls at that boundary. Fresh sandboxes can isolate a proposed fix, while tests and recordings are meant to make review less dependent on trusting a summary.
The scorecard is also a useful design choice: teams can inspect cost per session and model mix rather than assume that automation is cheaper. But observability is not evidence of savings, and an illustrative workflow cannot establish that agents reliably diagnose real failures or reduce review effort. A team considering it should treat the CI pilot as an evaluation: compare the agent’s evidence, cost and human time against its current process, including cases where it stops to ask for help.
Limits and uncertainty
WakeVane labels the product alpha, and its site says the first real run on a public repository has not yet been published. The workflow shown is illustrative, and the company says results depend on the task and configuration. No independent measurements establish fix accuracy, time saved, review burden or cost advantage. The scorecard describes metrics the product can expose, not results from a disclosed deployment. The site names 12 sandbox providers but does not identify them in the described material, and it does not state a price. Compatibility and behavior across the listed agents are also not independently checked here.
Skill to build now
Run the suggested CI-failure pilot on a low-risk repository for one week. Record each trigger, whether the agent correctly diagnoses the failure or stops to ask, the tests and evidence it returns, reviewer time, and session cost. Compare those results with the team’s normal diagnosis process before granting broader permissions or unattended scope.
Reader takeaway
WakeVane packages event-triggered coding-agent work around Omnigent, with cloud sandboxes, policy controls and session-cost visibility. That is a practical setup to test, not proof that autonomous fixes are ready to trust: the first public real-run evidence has not yet appeared. Evaluate it by the quality of returned evidence and the full review cost, not by the promise of work happening while you are away.