Doppler Finally Ships On-Prem — It's Still Not Infisical: What Self-Hosting Secrets Actually Costs

In 2020, a Hacker News commenter summed up the objection that followed Doppler for six straight years: "You have to give Doppler your secrets which is absolutely crazy. Is there a self-hosted [alternative]?" There wasn't. Doppler stayed closed-source and cloud-only no matter how good its integrations got, which is exactly why Infisical — MIT-licensed, self-hostable from day one — picked up every team that data residency or an audit requirement ruled out of Doppler entirely.

That changed on June 8, 2026, when Doppler announced On-Prem: the platform, packaged to run entirely inside a customer's own infrastructure. CEO Amber Britton framed it as following customers' AI workloads into constrained, regulated environments rather than asking them to keep trusting Doppler's cloud. It's a real product, not vaporware, and it's a direct answer to six years of the exact complaint above.

Here's the verdict up front: it doesn't close the gap. Not because Doppler On-Prem doesn't work, but because "we now offer on-prem" and "MIT-licensed, free to self-host, unlimited identities" are two different value propositions wearing the same marketing word. The rest of this post is the concrete cost comparison — infrastructure dollars, ongoing ops hours, and a pricing-model mismatch that flips the "self-hosting is cheaper" assumption on its head faster than you'd expect.

What "Doppler On-Prem" Actually Ships​

Doppler On-Prem is gated to the Enterprise tier only. Doppler's own pricing page lists three tiers: Developer (free for 3 users, then $8/user up to a 25-user cap, no SSO/RBAC/on-prem), Team ($21/user/month, up to 500 users, adds SAML SSO and RBAC — still no on-prem), and Enterprise, which is the only tier that includes on-prem or cloud deployment plus dynamic secrets and Enterprise Key Management. Enterprise pricing isn't published; industry trackers estimate roughly $30–40/user/month at 100+ seats, but that's a third-party estimate, not a quoted rate — you get a number by talking to sales.

Two things follow from that gating. First, on-prem availability doesn't change the trust model, only the network boundary. Doppler is still closed-source: you install a packaged binary, you don't get the source, and you can't fork it if Doppler discontinues the product, changes its enterprise pricing, or gets acquired. You're trusting the vendor on-prem exactly as much as you trusted them in the cloud — you've just moved the blast radius of that trust onto your own hardware. Second, it's priced for organizations already committed to a five- or six-figure annual contract. A small team self-hosting a PaaS to avoid exactly that kind of vendor dependency isn't the Enterprise-tier buyer this product was built for.

What Self-Hosting Infisical Actually Costs — Infrastructure​

Infisical's core — the API server, CLI, SDKs, Kubernetes operator, agent, and dashboard — is MIT-licensed and free to self-host with unlimited identities. That's not a trial; the ee/ directory in the repo is the only piece under a separate commercial license, and it's clearly separated from everything else. So the honest cost of self-hosting starts with infrastructure, not a license fee.

Infisical's Helm chart defaults give a concrete number to work from: the app pod requests 350m CPU / 1000Mi memory, PostgreSQL 250m CPU / 512Mi memory, and Redis 100m CPU / 256Mi memory. Infisical's own docs put the floor at one node with 2 cores / 4GB RAM / 20GB disk for a small deployment, and recommend 3+ nodes at 4 cores / 8GB RAM / 50GB+ SSD each for production.

Priced against Hetzner's current shared-vCPU lineup (post the mid-2026 rate adjustment): a single CPX22-class box (roughly matching the 2-core/4GB floor) runs about €19.49/month ($21). Three CPX32-class boxes (4 core/8GB, matching the production recommendation) run about €35.49/month each ($38), or roughly $114/month for the full HA footprint. That's the entire infrastructure bill for a production-grade, self-hosted Infisical cluster — cheap enough that it's not the number that decides this comparison.

The Number the Infra Total Leaves Out: Ops Hours​

"Free to self-host" and "free to operate" aren't the same claim, and this is where the real question — what does this cost next to a self-hosted PaaS — gets real. Running Postgres and Redis in production, even a small instance of each, is recurring work regardless of how cheap the VM is:

• Backups and restore drills — WAL archiving or scheduled pg_dump, plus actually testing a restore, not just confirming a cron job ran. Skip this and Infisical is a single-node outage away from losing every secret it holds.
• Version upgrades — Infisical ships frequent releases as an actively developed MIT project; each one needs a changelog read and a migration test before it touches production, since a secrets store is the one service you don't get to find out is broken during a deploy.
• Redis/Postgres HA and failover testing — the Helm chart can provision both in-cluster, but "provisioned" isn't "verified to fail over cleanly."
• On-call — a self-hosted secrets store sits on the critical path for every tenant deploy on the PaaS it's feeding. When it's down, nothing can fetch a fresh secret, which means it needs to be inside the same on-call rotation as the fleet itself, not an afterthought.

None of this is exotic — it's the same maintenance tax every self-hosted stateful service carries. But it's real engineering time, not a one-time infrastructure line item, and a reasonable estimate for a small production instance once initial setup is done is somewhere in the 4–6 hours/month range: backup verification, a changelog skim and occasional upgrade, and periodic failover checks. At a loaded engineering rate of $75–100/hour, that's roughly $300–600/month in labor — which is a bigger number than the $114/month infrastructure bill it sits on top of, and the one "self-hosting is basically free" pitches tend to leave out.

The Catch: Doppler Bills Humans, Infisical Pro Bills Everything With a Pulse​

Here's where the pricing models actually diverge, and it's not in the direction "self-hosted and MIT-licensed" implies at first glance. Infisical's free Community Edition has no RBAC, no SSO beyond basic Google/GitHub, no secret versioning, and no automatic rotation — those all require a Pro license at $18/month per identity, where "identity" counts every machine account (CI/CD tokens, service accounts, deploy keys) alongside every human user, even on a fully self-hosted instance validated against Infisical's license server. Doppler's Team tier, by contrast, bills $21/month per human seat — service tokens and machine identities are free.

That asymmetry means the crossover point arrives almost immediately. Take a 10-person team already needing RBAC and SSO (so: Doppler Team vs. Infisical Pro, not the free tiers on either side):

Doppler Team's bill is flat at $210/month regardless of how many CI runners, deploy bots, or service accounts pull secrets, because it doesn't count them. Infisical Pro crosses Doppler's price the moment a 10-person team has more than about two machine identities — and any real CI/CD pipeline clears that within its first pull request. Self-hosting only pays off here if the team stays on Infisical's free Community Edition and skips Pro's RBAC/SSO/rotation entirely; the moment those features are needed, "self-hosted and MIT-licensed" stops meaning "cheaper than Doppler."

What a Self-Hosted PaaS Operator Should Actually Do​

The way out of that trap is to notice that a self-hosted PaaS already has answers to most of what Infisical Pro sells. If the platform's own git-based deploy flow already scopes who can push to which app and which environment, that's most of what RBAC is buying — Infisical's Free tier's project/environment scoping plus the platform's own access control covers a lot of ground without a Pro license. SSO is the harder gap to route around if the org already runs Okta or Google Workspace company-wide, but for a team that hasn't hit that requirement yet, Free-tier self-hosted Infisical genuinely is free, not "free until you need one obvious feature."

The ops-hours number shrinks the same way. A self-hosted PaaS built on Cluster API is very likely already running Postgres and Redis for its own control plane, and already has backup automation, upgrade cadence, and on-call coverage for exactly those two dependencies. Running Infisical's Postgres and Redis alongside that existing stack — rather than standing up a fresh, isolated instance of each — turns most of the 4–6 hours/month estimate above into marginal cost on top of work the platform is already doing, not a new line item. That's the actual argument for self-hosting: not that it's magically free, but that a platform whose entire pitch is "own your infrastructure" already carries the fixed costs that make the secrets layer cheap to add, in a way a team standing up Infisical from scratch specifically to avoid Doppler wouldn't get for free.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. If you're already running Postgres and Redis for your own fleet, self-hosting your secrets layer next to it costs a lot less than standing it up in isolation. Star the repo on GitHub or deploy your first app today.

Sources​

• Doppler Brings Secrets Management On-prem for Full Control
• We're excited to announce Doppler On-prem - Doppler Community
• Doppler Pricing
• Infisical Pricing
• Infisical Enterprise Edition licensing
• infisical/LICENSE
• Kubernetes via Helm Chart - Infisical
• Infisical vs Doppler comparison
• Hetzner Cloud regular-performance pricing
• You have to give Doppler your secrets - Hacker News.