PA·dev vs EverOS
Same audience, different category. Here's the honest, technical breakdown — where EverOS wins, where PA·dev does, and which one fits what you're building.
Credit where it's due.
EverOS is a serious piece of work. It's an open-source, local-first memory operating system for agents — over 7,000 GitHub stars, a published research paper behind the design, and a best-in-class 93% on the LoCoMo long-conversation benchmark, measured head-to-head against Mem0, Zep, MemOS, and MemU. It runs entirely on your own machine — SQLite, an embedded LanceDB index, and Markdown as the source of truth — under Apache-2.0, with demos wired into Claude Code, OpenClaw, and the TEN framework. If you care about memory and you care about owning the whole stack, the EverMind team has earned the look.
So this isn't a takedown. EverOS and PA·dev sell to the same people — developers building agents — but they solve different halves of the problem. EverOS makes an agent remember better than almost anything else out there. PA·dev makes an agent act on what it knows: plan the work, get your approval, and execute inside your tools. The rest of this page is the dimension-by-dimension version of that split, and we hand EverOS the rounds it wins.
Memory-recall benchmarks
→ EverOS wins
EverOS is built to win this, and it does. 93% on LoCoMo, best-in-class against Mem0, Zep, MemOS, and MemU on the same datasets and answer model, with hybrid retrieval — BM25, vector, and scalar filter in a single query — and a research paper behind the design. If your product lives or dies on how much an agent recalls across long, noisy histories, this is the number to chase, and EverOS chases it harder than we do.
PA·dev's memory is good enough to run a business on: three tiers, privacy enforced server-side, recall that holds across every client that reads the brain. But we don't publish a LoCoMo score, because recall isn't the wedge we optimize for. If yours is, take theirs.
Infrastructure to run
→ PA·dev wins
EverOS is local-first and refreshingly light to stand up: SQLite, an embedded LanceDB index, and Markdown files, with no MongoDB, Elasticsearch, or Redis in sight. For a solo dev on a laptop, that's a real convenience.
But it's still a runtime you run and operate — on your machine for one user, on a server you host, scale, and back up the moment you have many. PA·dev is hosted. One key, nothing to provision, nothing to keep alive at 3am. The win here isn't “fewer databases” — EverOS already kept that light — it's not operating the memory layer at all.
What the agent does with the memory
→ PA·dev wins
This is the real line between the two. EverOS remembers — that's the whole job, and it does it well. PA·dev remembers and then acts.
You hand it a goal, it scaffolds a Project → Milestones → Tasks plan, you approve it, and a dispatcher fires a sub-agent at each task. Connectors are write-side: send the email, create the Stripe payment link, open the GitHub commit, trigger the Vercel deploy — each one staged to an approval gate before anything leaves the building. EverOS feeds an agent context; PA·dev is the agent's hands.
Open-source surface
→ EverOS wins
EverOS is open all the way down. The whole stack is Apache-2.0, so you can read every line, fork it, run it offline, and never depend on anyone's uptime.
PA·dev is hosted, and the only piece we open-source is the MCP wrapper, under MIT — enough to read the client and mock the API before you pay, but not the runtime itself. If self-hostable, auditable, no-vendor source is a hard requirement, EverOS is the honest answer and we're not going to pretend otherwise.
Multi-tenant SaaS / commercial readiness
→ PA·dev wins
PA·dev was built to sit under a product you sell. Multi-tenant out of the box, per-row privacy zones enforced server-side, pa_live_ API keys, bring your own model through one dispatcher, usage you can meter and bill.
EverOS is a single-tenant local runtime — ideal for one developer or one research setup, but standing up multi-user isolation, auth, and billing is work you'd own. If you're shipping agents to paying customers, PA·dev hands you the commercial scaffolding; if you're running your own agents, you may not need it.
EverOS figures — star count, LoCoMo score, license, architecture — are from the public EverMind-AI/EverOS repository and the EverMind benchmark writeups as of June 2026. If they ship a number we got wrong, the fix is one commit away.
Which one is right for you.
The honest answer is that these aren't the same product. If you're a researcher or an engineer benchmarking memory — chasing LoCoMo, comparing retrieval strategies, running everything local and open — EverOS is the better tool, and it's not close.
If you're shipping a product where the agent has to act inside your customer's tools — plan the work, get approval, send the email, make the charge — that's PA·dev. Same audience of agent builders, two different jobs. Pick the one that matches the job you actually have.