PermaShip

PermaShip's core claim is that judgment and execution should be separate layers, and it is the cleanest implementation of that idea in the autonomous coding space right now. The execution layer is a roster of domain-specialized agents — security, reliability, performance, cost, test coverage, CI health, frontend quality, product alignment — that scan repos continuously rather than waiting for a prompt. The judgment layer is Nexus, configured with your engineering philosophy and risk tolerance, and it holds a hard monopoly: only Nexus can create a ticket. That single constraint is what makes autonomous mode defensible. Most autonomous-agent products ask you to trust the agent that writes the code to also decide whether the code should be written. PermaShip splits those roles, and in March 2026 it open-sourced Nexus so the approval logic is inspectable rather than asserted. Where it fits: small teams that need senior-engineer output without the headcount. The homepage case study is concrete — a CTO connected a main repo on a Friday and had 14 PRs ready Monday morning (dependency updates, a missed failing test, two security flags), approving 11 before standup. Dev shops and agencies running a portfolio of repos get the most obvious leverage, because the work that never makes it into an SOW is exactly the work an always-on scanner finds. Where it doesn't: teams whose policy requires human line-by-line review of every merge, organizations where an AI-created ticket can't be reconciled with an existing compliance process, and repos with thin or unreliable CI, where a Nexus-approved PR is a real risk. Solo developers on a tight budget should be careful — build-from-scratch work at 15+ Compute drains the $20/mo Starter's 400 Compute quickly, and that is fine for a month of small fixes but not for a rewrite. Pricing is genuinely execution-shaped rather than seat-shaped: unlimited projects and unlimited seats on every tier, a monthly Compute budget, and per-ticket consumption of 1–3 Compute for a small fix or test, 5–15 for a feature or improvement, and 15+ for build-from-scratch or large refactors. The tier ladder buys throughput more than features — concurrent jobs go 1 → 5 → 20, retention 30 → 90 → 365 days, integrations 5 → 15 → unlimited, runner memory 4 → 8 → 16 GB, and Custom Docker starts at Team. Nexus governance and supervised/autonomous mode are on all plans, including Starter. The honest gaps: the /docs page says guides, API references, and integration docs are still being written, so expect to learn largely by doing. The publicly documented integration surface in this scrape is GitHub, and the integrations page wasn't reached on this run, so treat that as the confirmed floor rather than a ceiling. Full autonomy is a cultural decision as much as a technical one — the mechanism is sounder than most, but you are still deciding how much of your main branch to hand an agent.