Retool Performance: Benchmarks, Latency & Limits 2026
Retool is an internal-tools platform, not a web host, so performance is workflow throughput and plan limits. No published SLA, only status-page 90-day actuals.
Retool Performance verdict
Retool is an internal-tools platform on AWS, not a website host, so performance means workflow throughput and plan limits, not TTFB.
A non-production Docker instance handles roughly 5 to 10 workflow QPS, and production Kubernetes scales beyond it. There is no published SLA, only status-page 90-day actuals, with the web app at 99.05% and workflows at 99.97%.
Size Retool by seats and workflow runs, not servers. Free suits under 5 users and light workflows, with Team or Business as builders and runs grow. For heavy workflow throughput or data control, self-host on Kubernetes, since Docker is non-production only at 64 GiB and 8 vCPU minimum. With no published SLA, hold Retool to its status-page actuals and lean on the SOC 2 and ISO audits for assurance. Plan bring-your-own-key for AI on self-hosted, because the managed proxy times out at 29 seconds.
- Retool publishes no contractual SLA. The only reliability evidence is the status page's 90-day trailing actuals, with the web app and APIs at 99.05% and infrastructure at 99.11%.
- There is no independent third-party speed benchmark, since Retool is a platform not a website, so the independent signals are the SOC 2 and ISO audits, a pentest, and public CVEs. Workflow throughput of 5 to 10 QPS is a non-production Docker baseline, and production Kubernetes scales further without a published figure.
- Self-hosted AI requires bring-your-own-key, because the managed key proxy has a hard 29-second timeout that agent operations regularly exceed.
- Workflow throughput (Docker)
- ~5 to 10 QPS
- Published SLA
- None (status actuals)
- Web app uptime (90-day)
- 99.05%
- Independent audits
- SOC 2 Type 2 + ISO 27001
- Self-hosted minimum
- 64 GiB / 8 vCPU
This page covers how Retool scales under load and what reliability it can show. Region coverage and residency live on their own pages.
Browse the full Retool plan catalog
| Plan | Builder seat | Internal user | Workflow runs/mo | AI credits/mo | Modules |
|---|---|---|---|---|---|
| Free | $0 | $0 | 500 | 250 | Up to 5 |
| TeamPopular | $10 / seat | $5 / seat | 5,000 | 1,000 | Up to 5 |
| Business | $50 / seat | $15 / seat | 5,000 | 3,000 | Unlimited |
| Enterprise | Custom | Custom | Custom | 10,000 | Unlimited |
Cloud, billed annually (20% savings vs monthly). A builder edited an app or workflow in the cycle; an internal user did not. Business adds external-user pricing ($4 to $8/user) and portals; Enterprise adds SSO, source control and self-hosting.
Intro prices are first-term promotional rates. Cloud is billed annually (20% off monthly); pricing is per seat, not per server. Full catalog from Retool; verify live pricing before buying.
Throughput, timeouts and processing limits
| Dimension | Value | Scope | Source |
|---|---|---|---|
| Workflow throughput (Docker) | ~5 to 10 QPS | Non-production Docker baseline | Retool docs |
| Production throughput | Scales beyond 10 QPS | Kubernetes production | Retool docs |
| Agent key-proxy timeout | 29 seconds (hard) | Self-hosted with managed AI keys | Retool docs |
| Maintenance window | Under 15 min, off-hours | Retool Cloud | Retool docs |
| Multi-tenant release cadence | Multiple times/week | Retool Cloud multi-tenant | Retool docs |
Plan limits and resource ceilings
| Limit | Free | Team / Business | Notes |
|---|---|---|---|
| Workflow runs/mo | 500 | 5,000 | Enterprise custom; Temporal schedules and queues runs |
| AI credits/mo | 250 | 1,000 / 3,000 | Enterprise 10,000; credits renew each cycle, no rollover |
| Agent free hours/mo | 20 | 20 | Billed hourly by wall-clock; idle wait excluded |
| Database / file capacity | 5 GB / 5 GB | Not listed | RetoolDB and file storage on Free; higher tiers unlisted |
| Web and mobile apps | Unlimited | Unlimited | Unlimited on every plan |
| Modules | Up to 5 | 5 / Unlimited | Unlimited on Business and Enterprise |
Scaling model across deployment options
| Aspect | Entry | Ceiling | Scales by |
|---|---|---|---|
| Self-hosted compute | Docker 64 GiB / 8 vCPU | EKS multi-pod (prod) | Vertical VM to horizontal Kubernetes |
| Production pods | Single Docker container | API + workers + executors + sandboxes | Separate pods per role on EKS |
| Workflow runs | 500/mo (Free) | 5,000/mo (Team/Business), Custom | Plan tier; Temporal orchestration |
| Environments | Staging (Team) | Unlimited (Business/Enterprise) | Multi-environment promotion |
| Tenancy | Multi-tenant Cloud | Single-tenant / self-managed | Shared SaaS to dedicated/self-hosted |
| Source control | Enterprise feature | Git-as-code, 5 providers | GitHub/GitLab/Bitbucket/CodeCommit/Azure |
Retool reliability and architecture
- Retool Cloud runs on Amazon Web Services with infrastructure that is fault-tolerant for failures of individual servers or entire data centers, with customer data stored redundantly in multiple locations
- There is no published SLA percentage; the status page reports 90-day trailing actuals, Web app and APIs 99.05%, Infrastructure 99.11%, Resource Queries 99.74%, and Workflows and RetoolDB 99.97%
- Backups are automated nightly and retained for 7 days, tested at least every 90 days, with disaster recovery regularly tested and a 24-hour on-call operations team
- For external data sources Retool is a query proxy and does not store customer data, only app definitions, permissions, audit logs and workflow state, which reduces the data-loss blast radius
- The platform uses a multi-tenant architecture at both the platform and infrastructure layers, designed to segregate and restrict access between tenants
- On self-hosted deployments no Retool systems store customer data and no Retool personnel have technical or logical access to it, so reliability and data control shift entirely to the customer's infrastructure
Independent verification and findings
- Retool holds a SOC 2 Type 2 report, an independent third-party audit of its security and availability controls, available in its trust center
- Retool is certified to ISO/IEC 27001:2022, an independent information-security management certification, with the document available in the trust center
- An independent penetration-test report is available in the trust center, alongside a SIG Lite self-assessment, subprocessor list and data-processing agreement
- Public vulnerability disclosures provide an independent integrity signal: CVE-2025-47424, a host-header injection affecting self-hosted instances without BASE_DOMAIN, was fixed in 3.196.0 and later (CVSS 7.1)
- Retool states it will at a minimum align with prevailing industry standards such as SOC 2 Type 2, with continuous automated scanning, 24/7 alerting and logs reviewed at least every 90 days
Retool Performance FAQ
How does Retool perform, and is it benchmarked?
Retool is a platform, not a website host, so there is no TTFB or third-party speed benchmark. The performance numbers are workflow throughput, roughly 5 to 10 QPS on a non-production Docker instance, with production Kubernetes scaling beyond and no specific figure published. The managed AI key proxy has a 29-second timeout. The independent signals are Retool's SOC 2 Type 2 and ISO 27001 audits, an independent pentest, public CVEs, plus the status page's 90-day actuals.
What is Retool's SLA and actual uptime?
Retool publishes no contractual SLA percentage on its public pages. The status page reports 90-day trailing actuals by component: the web application and APIs at 99.05%, infrastructure at 99.11%, resource queries at 99.74%, workflows and RetoolDB at 99.97%. License, email, source-control and audit-trail components showed 100%. With no SLA, hold Retool to these published actuals rather than a guarantee.
How do Retool plans and limits work?
Plans are per seat, split into builder and internal user, billed annually with a 20% saving. Free allows up to 5 users, 500 workflow runs and 250 AI credits a month, with 5 GB of database and file storage. Team, at $10 per builder and $5 per user, raises runs to 5,000 and credits to 1,000. Business, at $50 and $15, reaches 3,000 credits, unlimited modules and external-user pricing. Web and mobile apps are unlimited on every plan, and agents include 20 free hours a month.
What are Retool's self-hosting resource requirements?
A Docker self-hosted instance needs Ubuntu 22.04 or newer on x86, with a 64 GiB RAM, 8 vCPU and 60 GiB storage minimum, but Docker is for non-production only. Production must run on Kubernetes with Helm, EKS on AWS, which splits into separate API, workflow worker, backend, code-executor and agent-sandbox pods. Self-hosting is an Enterprise feature.
How does Retool scale under load?
On Cloud, you scale by plan tier and seats, and the multi-tenant infrastructure updates several times a week. Self-hosted, you scale horizontally on Kubernetes with EKS, where the API, workflow and executor pods scale independently, and Temporal orchestrates workflow runs. The published throughput is a 5 to 10 QPS Docker baseline, with production Kubernetes scaling further, though Retool publishes no specific production QPS.
Does Retool store my data, and how is it backed up?
For external data sources, Retool is a query proxy and does not store your data, only app definitions, permissions, audit logs, workflow state, cached temporarily for a duration you set. On self-hosted, no Retool systems store customer data at all. Backups are nightly with a 7-day retention, tested every 90 days, with disaster recovery regularly tested and a 24-hour on-call team.
Sources & verification
| Source | What was checked | Last checked |
|---|---|---|
| Retool Official | Official product page | July 10, 2026 |
| NIST | NIST framework requirements | July 10, 2026 |
| Retool Developer docs | Tutorials Docker | July 10, 2026 |
| Retool Developer docs | Tutorials Kubernetes | July 10, 2026 |
| Retool Developer docs | Security and compliance | July 10, 2026 |
| Retool Pricing | Pricing and plans | July 10, 2026 |
| Retool Status | Product documentation | July 10, 2026 |
Every fact on this Retool page is tied to a named source and a verification date. Freshness-sensitive figures trace to the sources above; verify against the vendor before relying on them.
Explore Retool
Every page on Retool in one place, you are on performance.