Debugging an incident on Cloudflare has always meant knowing which product owns which signal. A spike in 5xx responses could come from a Worker, from your origin, or from Cloudflare failing to connect to your origin regionally — and each possibility lived in a different tool with a different query model. On October 2, 2026, Cloudflare’s blog post announced a bundle of changes meant to collapse that fragmentation into a single platform, with one pricing model behind it.
What actually shipped
The headline items: a unified Logs home that merges Workers Observability with Log Explorer, letting you switch between datasets like HTTP events, firewall events, Workers, Containers, R2, and AI Gateway without leaving the investigation. Cloudflare Traces enters open beta, giving a request-level view of security rules, cache decisions, routing, Workers, and origin handling, with a baseline sampling rate plus Trace Rules to capture specific traffic at higher rates during an incident.
On top of that: custom alerts on any dataset the SQL API supports, domain analytics consolidated in one place with 30 days of retention on every plan, custom dashboards mixing logs, traces, and security events, and Logpush — previously Enterprise-only — now on all self-serve plans, with the SQL-based Transformers feature generally available for filtering and redacting events before export.
The part that matters for agent builders
Buried in the middle of the announcement is the most interesting shift: a unified SQL API, now in beta, designed explicitly for agents as well as people. Instead of integrating separately with Workers logs, Containers security events, HTTP request logs, and analytics, both humans and agents query everything through one SQL dialect, one authentication model, and one API.
Cloudflare supports this with its cf CLI, an Observability MCP server, and published dataset schemas with example queries so an agent can discover and build queries on its own. There’s also a native Workers binding, so a Worker can query Analytics Engine data to meter customer usage or power billing workflows without a separate API client. The blog post is unusually explicit about the motivation: an agent can only close the loop between a change and its outcome if it can query what happened, identify the failure, and verify the fix.
This is the same pattern we saw when observability-adjacent capabilities became part of the API surface — infrastructure that used to be dashboard-only is becoming callable, which is exactly what you need if agents are going to operate software rather than just write it. Cloudflare also built around OpenTelemetry, W3C Trace Context, and SQL rather than proprietary formats, which keeps that context portable to the rest of your stack.
Pricing: simpler, but check your December math
The unified pricing takes effect December 1, 2026, across all plans (upon renewal for Enterprise). It covers Workers, Containers, AI Gateway logs, and tracing data, and it’s volume-based rather than event-count-based, since log and trace sizes vary dramatically.
Free plans get 0.5 GB of ingestion per day with 7 days retention. Paid plans include 50 GB of ingestion and 10 GB-month of storage per billing cycle, then $0.25 per GB ingested and $0.10 per GB-month stored, with retention up to one year coming soon. Logpush and Transformers move to usage-based pricing with monthly allowances — 25 GB free for exports, $0.03 per GB to Cloudflare destinations, $0.10 per GB to external ones, $0.04 per GB for Transformers.
If you’re currently paying per-event for Workers logs, run the numbers before renewal; a chatty Worker with verbose tracing could get cheaper or more expensive depending on your event sizes.
What’s still ahead
Cross-dataset querying in a single query is listed as coming soon, as is the one-year retention, OpenTelemetry API support in Workers for adding span attributes, and metrics export to OpenTelemetry-compatible destinations. Cloudflare says more products and datasets will join the platform over the coming months.
Practical next step if you run agents on Cloudflare: point one at the cf CLI or the Observability MCP server with the schema docs, and see whether it can answer “what changed after the last deploy” end to end. That’s the loop these changes are betting on.