← Back to Blog Jul 15, 2026 PR —

What's New in v0.12.0 — And How to Get Started

This post is different from the others in this series. Most entries here are engineering diary — the "why" and "how" behind a specific PR, written for people already using synlynk. This one is for you if you're deciding whether to try it, or if you installed it a few months ago and haven't kept up.

The one-sentence version

synlynk is a CLI that keeps your AI coding tools — Claude, Codex, Gemini (Agy), and Grok — in sync with a shared project state, so you can dispatch work to whichever agent fits, track what it cost, and trust that the numbers you're seeing are real.

If you're new here

Three commands get you running:

pipx install git+https://github.com/nikhilsoman/synlynk
synlynk init --wizard
synlynk exec claude

init --wizard walks you through an 8-screen setup: which AI CLIs you have installed, solo vs. team mode, and whether you want GitHub issue/project linking. exec wraps any AI CLI call with the current project context — active tasks, recent decisions, budget status — so the agent starts every session already caught up.

From there, the two commands you'll use daily are synlynk dispatch <agent> --task "..." to hand off a background job, and synlynk jobs --watch to see it land. Full details are in the Quick Start Guide and Command Reference.

If you've been away for a bit — what changed

v0.12.0 shipped 71 PRs since v0.11.0 (2026-07-05 → 2026-07-15), and the theme across all of them was trust: making sure dispatched work actually lands, routing is based on real signal instead of a coin flip, and every dollar synlynk reports is either measured or clearly labeled as a guess.

Five things worth knowing about if you haven't upgraded recently:

  • Dispatched jobs finish themselves. synlynk used to leave commit/push/PR steps to the agent, and agents didn't reliably do them. Now synlynk does it — the moment a job's work is verifiably complete, it stages, commits, pushes, and opens the PR for you.
  • Routing got smarter. Dispatch used to be close to first-match. Now it scores agents on capability, checks quota headroom across five time windows, and tie-breaks on cost — and if you have a backlog, synlynk schedule --execute will clear it unattended.
  • Costs are provably real. Every number in synlynk status and the Vizor dashboard (synlynk viz) is now tagged as either a structurally-sourced measurement or a visibly-flagged estimate — no more silent guesses dressed up as facts. If you're logging cost for work synlynk didn't wrap directly (a native session, a manual fix), synlynk cost log records it properly.
  • synlynk status has a new RATES line showing when the pricing table was last refreshed, with a warning if it's gone stale.

Getting current

synlynk upgrade
synlynk doctor

doctor runs a health check after the upgrade — worth running even if upgrade reports success, since it also validates your .synlynk/config.json and agent CLI reachability.

Where to go next

If you want the engineering detail behind any of this — the actual diffs, the design decisions, the dead ends — the rest of this series has it, starting with post 64 for the full v0.12.0 release notes.

← Back to all posts