Web Search API is now available in beta. Web Search API lets your AI agents and applications search the Internet and ground their responses in live information, instead of guessing URLs or relying on a model's training cutoff.
At launch, you can choose between three search providers: Ceramic.ai, Exa, and Linkup. All three support Zero Data Retention for requests made through Cloudflare, and all have committed to Cloudflare's verified bot crawling standards.
Web Search API runs through AI Gateway, so search requests appear in your gateway logs and are billed to your AI Gateway credits at each provider's list API price, with no additional markup. You can also bring your own provider API key.
Call Web Search API with the REST API:
curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/websearch/ \ --request POST \ --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \ --header "Content-Type: application/json" \ --data '{ "query": "What are some fun things to do in Salt Lake City as fall approaches?", "provider": "ceramic", "limit": 5, "options": { "gateway": { "id": "default" } } }'
Or from a Worker with the AI binding:
const response = await env.AI.websearch({ gatewayId: "default", query: "What are some fun things to do in Salt Lake City as fall approaches?", provider: "exa", limit: 5,});const results = await response.json();
The strict service token authentication setting applies consistent behavior to requests made with service tokens. When the setting is on for a Zero Trust organization, Access handles requests with service token headers as follows:
If authentication or authorization fails, Access always returns 401 or 403 instead of redirecting the client to the login page with 302.
Only Service Auth policies can authorize the request. Access ignores Allow policies and any CF_Authorization cookie sent with the request.
Access does not return a CF_Authorization cookie to the client after successful authentication. Subsequent requests should continue to use service token headers.
Zero Trust organizations created on or after October 5, 2026 have strict service token authentication turned on by default and cannot turn it off. Cloudflare recommends that existing organizations turn it on as well.
Organizations created before October 5, 2026 can configure the setting in the dashboard or through the API.
The Agents SDK now provides first-class support for building agents using the Pi harness. You can build long-running agents using the combination of Pi 1.0 ↗︎, Pi Durable ↗︎, and the new PiHarness class that the Cloudflare Agents SDK provides, ensuring your agent's work is durably persisted, even if interrupted mid-turn.
Built with Earendil ↗︎, this integration is our first step toward first-class support for third-party agent harnesses on Cloudflare.
PiHarness is a new "Lifecycle capability" provided by the Cloudflare Agents SDK. Pi Durable provides the agent harness and the Lifecycle is responsible for keeping the agent running in the Durable Object. The Lifecycle is a core concept in the Agents SDK ensuring that long-running work can run in a Durable Object, surviving restarts, crashes, and network issues. We will share more on Lifecycle capabilities in the near future.
Install
npm i agents@latest @earendil-works/pi-durable @earendil-works/pi-ai
The agents/models/pi-ai entry point supports AI Gateway and Workers AI models, so you can get started with Cloudflare models right away or use your existing pi-ai provider.
Add tools with extensions
Both tools and system prompt sections are provided to the Pi Harness via extensions.
import { Type } from "@earendil-works/pi-ai";import { skills } from "agents/harness/pi";const WordCount = Type.Object({ text: Type.String() });const wordCount = { name: "word_count", description: "Count the words in a text.", parameters: WordCount, replay: "safe", async execute({ text }) { const words = text.split(/\s+/).filter(Boolean).length; return { content: [{ type: "text", text: String(words) }] }; },};// In the harness factory, before Harness.open():registry.install({ name: "editor", sections: [ { key: "preamble", render: () => "You are an editor.", tag: false }, ], tools: [wordCount],});registry.install(await skills(sources));
import { Type } from "@earendil-works/pi-ai";import type { ToolRegistration } from "@earendil-works/pi-durable";import { skills } from "agents/harness/pi";const WordCount = Type.Object({ text: Type.String() });const wordCount: ToolRegistration<typeof WordCount> = { name: "word_count", description: "Count the words in a text.", parameters: WordCount, replay: "safe", async execute({ text }) { const words = text.split(/\s+/).filter(Boolean).length; return { content: [{ type: "text", text: String(words) }] }; },};// In the harness factory, before Harness.open():registry.install({ name: "editor", sections: [ { key: "preamble", render: () => "You are an editor.", tag: false }, ], tools: [wordCount],});registry.install(await skills(sources));
For more information on creating and configuring extensions, refer to Extensions.
Every plan now gets at least 30 days of analytics data. Adaptive analytics datasets, such as HTTP requests, security events, and DNS analytics, retain at least 31 days of data for Free and Pro domains, and you can query up to 30 days in a single request. Previously, Free and Pro domains could see between 24 hours and 8 days of history depending on the dataset.
A full month of history lets you investigate an issue after it happens, compare today with the same day in previous weeks, and tell a one-time spike from a longer trend. The change applies in the Cloudflare dashboard, in Custom Dashboards, and through the GraphQL Analytics API.
Domain analytics also now live in one place. In the Cloudflare dashboard, select a domain and go to Analytics to see Traffic, Performance, Security, Cache, Origin, DNS, and Visitors as tabs that share one time range and one set of filters. Account-level analytics are under Observability > Analytics.
This change does not alter which datasets or fields your plan can access. Aggregated datasets, such as httpRequests1hGroups, keep their existing per-plan limits. To check the exact retention and query window for a zone or account, query the settings for each dataset.
You can now build Custom Dashboards charts from Workers Observability data. Two new datasets, Workers Observability — Logs and Workers Observability — Traces (OTel), let you chart Worker invocations, log levels, errors, CPU and wall time, span counts, and durations next to HTTP traffic, security events, and other analytics datasets.
This gives you one dashboard for an application that spans Cloudflare's network and your Workers. For example, you can put request volume, WAF blocks, and Worker error rates on the same view, filter all three by time range, and spot whether a spike in errors lines up with a change in traffic.
The datasets are available for every Worker in your account that has Workers Logs or Workers Traces turned on. Custom Dashboards also now allow up to 100 dashboards for every account.
Jurisdictions for Workers KV namespaces are now generally available. When you create a namespace, you can set a jurisdiction to make sure the namespace's data is only durably stored within that region. Jurisdictions can help you comply with data localization regulations such as GDPR or FedRAMP. Supported jurisdictions are eu, us, and fedramp.
A jurisdiction can only be set when a namespace is created, using the Cloudflare dashboard, Wrangler, the cf CLI, or the REST API, and cannot be added or changed afterwards.
Workers can still access a namespace restricted to a jurisdiction from anywhere in the world, and KV data can be cached outside the jurisdiction on Cloudflare's network. The jurisdiction only controls where the namespace's data is durably stored.
hash_in_range() is globally available for HTTP products on all plans. It hashes fields into an integer within a specified range. Use this result to select a portion of requests.
Use cf.random_seed to select approximately 10% of requests at random:
hash_in_range(0, 100, cf.random_seed) < 10
With Cloudflare for SaaS, use custom metadata to control rollout progression. Define rollout_pct as a custom key for each hostname. Set its value to an integer from 0 to 100. The expression selects approximately that percentage of requests:
You can now restrict who can access a Quick Tunnel. Use the new --allowed-mail flag in cloudflared to require visitors to authenticate with a one-time PIN sent to their email before they reach your local service.
Previously, anyone with a trycloudflare.com URL could access the service behind it. Protected Quick Tunnels let you share a local development server, webhook receiver, or demo with specific people without creating a Cloudflare account or configuring a domain.
You can allow:
A single email address: --allowed-mail alice@example.com
Multiple email addresses, by repeating the flag or using a comma-separated list: --allowed-mail 'alice@example.com,bob@example.com'
Every address on a domain: --allowed-mail '*@example.com'
Visitors do not need a Cloudflare account. Access ends for everyone when you stop the cloudflared process.
Clef is a decision model, in the same family as Typesafe's Jev ↗︎. Instead of generating text, it reads an input state and a set of typed questions, then returns a probability for every allowed answer. Your agent gets a structured decision it can act on immediately, for example: route the ticket, block the request, or escalate to a human. There is no free-form output to parse and no reasoning tokens to wait for.
Both models are hosted on Workers AI as Clef and Clef-flash. We are also open-sourcing the weights under the Apache 2.0 license on Hugging Face: Clef ↗︎ and Clef-flash ↗︎. Read the launch blog post ↗︎ for the full story, including how we trained them.
Clef is designed to be fast so decisions come back in milliseconds. Across our 43 benchmark runs, we achieved speeds where Clef is 2.5x faster than Jev at the median, and Clef-flash 13x faster.
Latency
Clef
Clef-flash
Jev
Median
209.3 ms
38.8 ms
524.1 ms
p95
238.6 ms
122.4 ms
536.0 ms
Hosting on Workers AI adds to that speed. Requests run on GPUs across Cloudflare's network, running close to your users, so the network round trip stays short. You can put Clef directly in the request path of your agent, then hand off to an LLM on Workers AI to take action.
Leading the benchmarks
Across 10 decision benchmarks, a Clef model scores highest on 7, ahead of Jev and other open decision models. A few highlights:
Benchmark
Clef
Clef-flash
Jev
BFCL (case exact)
98.47
98.76
95.75
BANKING77 (macro-F1)
94.20
90.93
79.74
CLINC150+OOS (macro-F1)
97.43
66.77
89.27
Home appliances (case exact)
82.95
97.73
52.27
On Typesafe's own workflow evals, Clef beats Jev in 3 of 4 areas: invoice processing, customer service, and security incidents. The full results are on the Hugging Face model card ↗︎.
Clef follows the System One API, so you can switch an existing Jev integration to Clef by changing the endpoint and model. Ask up to 64 questions per request, in three types:
noul: A yes/no question. Returns the probability that the answer is yes.
choice: Pick one option from a set you define. Returns the chosen option, a probability per option, and a confidence value.
score: Rate against an ordered rubric. Returns a probability-weighted score and a probability per level.
const response = await env.AI.run("@cf/cloudflare/clef", { model: "clef", state: "Checkout has been failing for every customer for the last hour.", questions: { urgent: { type: "noul", instructions: "Is this support request urgent?", }, team: { type: "choice", instructions: "Which team should handle this request?", criteria: { billing: "Payments, invoices, and refunds", technical: "Outages, errors, and configuration", sales: "Plans and upgrades", }, }, },});// response.answers.urgent.noul -> probability the request is urgent// response.answers.team.choice -> highest-probability team
const response = await env.AI.run("@cf/cloudflare/clef", { model: "clef", state: "Checkout has been failing for every customer for the last hour.", questions: { urgent: { type: "noul", instructions: "Is this support request urgent?", }, team: { type: "choice", instructions: "Which team should handle this request?", criteria: { billing: "Payments, invoices, and refunds", technical: "Outages, errors, and configuration", sales: "Plans and upgrades", }, }, },});// response.answers.urgent.noul -> probability the request is urgent// response.answers.team.choice -> highest-probability team
What you can build with decision models
Support triage: Decide whether a ticket is urgent and which team owns it, then route it without a human in the loop.
Threat intelligence: Classify a website by category. Paired with Browser Run, Clef fetched, rendered, and classified a domain in 2.2 seconds, compared to 4.7 seconds for gpt-oss-120b in the same workflow.
Trust and safety: Score user submissions against your own policy rubric and act on the probability.
Agent guardrails: Let an agent check "should I take this action?" in tens of milliseconds before calling a tool.
Visual classification: Pass up to four images alongside the state. Unlike text-only decision models, Clef has a vision encoder.
Get started
Use Clef through the Workers AI binding (env.AI.run()) or the REST API at /ai/run. You can also use AI Gateway with these endpoints.
@cloudflare/workers-oauth-provider ↗︎ is now v1, with a new split API. One Worker acts as the authorization server: it signs users in and issues tokens. Your MCP server acts as the resource server, and can run in another Worker. It validates each token with the authorization server over a Service Binding, without crossing the public Internet.
A migration skill ↗︎ ships in the npm package so a coding agent can perform the upgrade.
The split API
import { OAuthAuthorizationServer, OAuthResourceServer, insufficientScope,} from "@cloudflare/workers-oauth-provider";import { WorkerEntrypoint } from "cloudflare:workers";// auth-server Worker: signs users in and issues tokens for both MCP servers.const authorizationServer = new OAuthAuthorizationServer({ issuer: "https://auth.example.com", resources: [ "https://calendar.example.com/mcp", "https://drive.example.com/mcp", ], scopesSupported: ["calendar:read", "calendar:write", "offline_access"], clientIdMetadataDocumentEnabled: true,});export class AuthServer extends WorkerEntrypoint { fetch(request) { if (new URL(request.url).pathname === "/authorize") { return showConsent(request, this.env); } return authorizationServer.fetch(request, this.env, this.ctx); } validateToken(resource, token) { return authorizationServer.validateToken(resource, token, this.env); }}// calendar MCP Worker: checks tokens with AuthServer over a Service Binding.export const calendar = new OAuthResourceServer({ resourceMetadata: { resource: "https://calendar.example.com/mcp", authorization_servers: ["https://auth.example.com"], }, requiredScopes: ["calendar:read"], validateToken: (env) => env.AUTH_SERVER.validateToken, handler: { fetch(request, env, ctx) { if ( request.method === "POST" && !ctx.auth.scope.includes("calendar:write") ) { return insufficientScope(ctx.auth, ["calendar:read", "calendar:write"]); } return handleMcp(request, ctx.props); }, },});
import { OAuthAuthorizationServer, OAuthResourceServer, insufficientScope,} from "@cloudflare/workers-oauth-provider";import { WorkerEntrypoint } from "cloudflare:workers";// auth-server Worker: signs users in and issues tokens for both MCP servers.const authorizationServer = new OAuthAuthorizationServer<Env>({ issuer: "https://auth.example.com", resources: [ "https://calendar.example.com/mcp", "https://drive.example.com/mcp", ], scopesSupported: ["calendar:read", "calendar:write", "offline_access"], clientIdMetadataDocumentEnabled: true,});export class AuthServer extends WorkerEntrypoint<Env> { fetch(request: Request) { if (new URL(request.url).pathname === "/authorize") { return showConsent(request, this.env); } return authorizationServer.fetch(request, this.env, this.ctx); } validateToken(resource: string, token: string) { return authorizationServer.validateToken(resource, token, this.env); }}// calendar MCP Worker: checks tokens with AuthServer over a Service Binding.export const calendar = new OAuthResourceServer<Env, AuthProps>({ resourceMetadata: { resource: "https://calendar.example.com/mcp", authorization_servers: ["https://auth.example.com"], }, requiredScopes: ["calendar:read"], validateToken: (env) => env.AUTH_SERVER.validateToken, handler: { fetch(request, env, ctx) { if ( request.method === "POST" && !ctx.auth.scope.includes("calendar:write") ) { return insufficientScope(ctx.auth, ["calendar:read", "calendar:write"]); } return handleMcp(request, ctx.props); }, },});
In the example, env.AUTH_SERVER.validateToken is that Service Binding call. The calendar Worker needs no KV namespace of its own.
{ "name": "calendar-mcp", "main": "src/index.ts", // Set this to today's date "compatibility_date": "2026-10-02", "services": [ { "binding": "AUTH_SERVER", "service": "auth-server", "entrypoint": "AuthServer", }, ],}
name = "calendar-mcp"main = "src/index.ts"# Set this to today's datecompatibility_date = "2026-10-02"[[services]]binding = "AUTH_SERVER"service = "auth-server"entrypoint = "AuthServer"
OAuthResourceServer publishes the RFC 9728 ↗︎ protected resource metadata that MCP clients use to find your authorization server. It answers requests without a token with a 401 challenge that points to that metadata. It also rejects tokens issued for any other resource.
You can still use OAuthProvider as both the authorization server and the MCP server. For most 0.x deployments, the only required change is to add resourceMetadata: { resource }.
AI Search is now generally available. Usage-based billing begins on November 1, 2026, with included monthly ingestion, storage, semantic query, and full-text query usage. Cloudflare will send a reminder email the week before billing begins.
New AI Search instances use hybrid search by default. Hybrid search combines semantic vector retrieval with full-text matching. You can choose a different index method when you create an instance.
Workers AI embedding and reranking calls made by AI Search are included in AI Search pricing. These calls no longer appear on your Workers AI bill or in your AI Gateway logs. Generation, query rewriting, and external providers continue to use your account and gateway.
AI Search supports the @cf/qwen/qwen3-vl-embedding-2b and google-ai-studio/gemini-embedding-2 multimodal embedding models. Search and chat requests can include images through the REST API and public endpoint.
Refer to Supported models for the full list of embedding models.
OCR availability and increased file limits
Optical character recognition (OCR) is available on every account for scanned PDFs. Plain-text or code files and PDFs with OCR enabled can be up to 10 MiB. PDFs without OCR and other supported formats remain limited to 4 MiB.
When you create an AI Search instance, the type field is optional. AI Search infers a website source from an HTTP or HTTPS URL, or an R2 source from an existing bucket name.
Artifacts, Cloudflare's versioned file system that speaks Git, is now in open beta. Artifacts is built for scale, so you can create a repository per project, user, session, or task.
With Artifacts, you can:
Deploy repositories to Workers — Connect an Artifacts repository through Workers Builds. Pushes to the production branch deploy the updated Worker, while other branches create or update Worker Previews.
Programmatically manage repositories — Use an Artifacts binding from a Worker to create or fork repos, inspect files and commits, read files by path, and issue repo-scoped Git tokens.
React to repository changes — Subscribe to events when a repository is created, imported, forked, deleted, pushed to, cloned, or fetched.
Control where repository data is stored — Choose to store and process your data in the US or EU.
Monitor repository usage — View total operations, pulls, pushes, errors, and error rates in the Cloudflare dashboard or via API for analytics.
Artifacts is available for customers on the Workers Paid plan. Cloudflare will begin billing for Artifacts on October 14, 2026.
Build the next GitHub on Cloudflare
We are hosting a competition to see who can build the next GitHub on Cloudflare using Workers and Artifacts.
Apply today ↗︎ — submissions are open until October 14, 2026.
The first-place team will receive $25,000 in Cloudflare credits. The top three teams will be flown to San Francisco to present what they built at Cloudflare Connect.
The higher per-stream limit gives high-volume application events, telemetry, and logs more room to grow without splitting ingestion across streams solely to stay within the previous limit.
For the full list of stream, sink, and pipeline limits, refer to Basin Pipelines limits.
Basin, formerly the Cloudflare Data Platform, is now generally available. Basin brings an end-to-end analytics platform to the Developer Platform, enabling you to collect data from a variety of sources, such as apps, infrastructure, devices, and other Cloudflare services, then query it to answer analytical questions.
Basin Pipelines
Basin Pipelines, formerly Cloudflare Pipelines, ingests events from Workers, HTTP endpoints, and Cloudflare Logpush. It transforms events with SQL and ingests them into Iceberg tables or files on R2. With Basin Pipelines you can:
Ingest application and device events through HTTP endpoints or Workers bindings.
Filter and reshape Cloudflare logs before storing them as Iceberg tables, Parquet, or JSON.
Catch schema mismatches with typed bindings and investigate dropped events in the dashboard.
Basin Catalog
Basin Catalog, formerly R2 Data Catalog, manages and automatically maintains Apache Iceberg tables ↗︎ to keep them fast, cost-efficient, and accessible to any compatible query engine. With Basin Catalog you can:
Connect DuckDB, Spark, Snowflake, or PyIceberg to the same tables.
Maintain growing tables with compaction, snapshot expiration, and manifest optimization.
Share analytical data across tools and clouds without paying egress fees.
Basin SQL
Basin SQL, formerly R2 SQL, is a serverless, distributed SQL engine for querying large Apache Iceberg tables in Basin Catalog without managing or scaling compute. With Basin SQL you can:
Summarize and rank data with standard and approximate aggregates, grouping sets, and window functions.
Combine and inspect datasets with joins, subqueries, common table expressions, set operations, schema discovery, and EXPLAIN.
Transform strings, timestamps, JSON, and complex values with more than 190 functions.
Get started
To get started with creating an end-to-end data pipeline, run:
You can now tag targets using only the Zero Trust Write API token permission. Previously, tagging targets through the API required both Zero Trust Write and Tag Write permissions on the API token.
Durable Objects remain active while handling a request from a connected client. This change applies when no client is connected, such as when an agent continues a submitted job after its client disconnects.
This behavior is the default for Workers with a compatibility date of 2026-10-01 or later. To use it with an earlier date, add the durable_object_io_tasks_prevent_eviction compatibility flag. To opt out, add the durable_object_io_tasks_do_not_prevent_eviction flag.
Pending service binding requests now keep Durable Objects running while they wait for a response. Pending calls to another Durable Object through remote procedure call (RPC) or fetch(), as well as this.ctx.container.monitor(), now also keep the Durable Object running.
Promises passed to this.ctx.waitUntil() and pending setTimeout() and setInterval() timers also receive this protection.
Previously, Cloudflare could shut down an idle Durable Object while one of these operations remained pending without a connected client. This could stop unfinished work.
This change helps you run long-running tasks such as agents. An agent can call tools through service bindings, coordinate with other Durable Objects, or wait for a container process without relying on the original client to remain connected.
Outbound fetch() requests to external services, TCP sockets, and outbound WebSockets already keep Durable Objects running.
Each pending operation prevents idle shutdown for up to 15 minutes. Starting another one later can extend the Durable Object's time in memory. The limit applies to each operation, not to the total time in memory.
Duration charges continue while an operation prevents eviction.
Account API token creation is no longer limited to Super Administrators. Members with the API Token Provisioning role can now create Account API tokens via the Dashboard, API, Terraform, or CF CLI, making it easier for developers and platform teams to provision credentials without depending on a Super Administrator for Account API Token Provisioning.
What's new
Delegated creation: Members with the API Token Provisioning role can create Account API tokens from the dashboard. Administrators can grant this role through the dashboard, API, or Terraform.
OAuth support for token creation: OAuth clients that request the account_api_tokens:create scope, starting with Cloudflare CLI, can create Account API tokens.
Account API token permissions limited to the creator’s access at creation time: Members can only create an Account API Token using the permissions they already have. For OAuth-created tokens, permissions are also limited to the scopes granted during authorization.
Creator attribution and visibility: Account API tokens now include creator metadata. Super Administrators and Administrators can view all Account API tokens in an account, while members with the API Token Provisioning role can only view tokens they created.
Cloudflare Rules expressions now support dynamic values on both sides of equality and ordering comparisons. You can compare request fields or function results with one another.
For example, compare the current request path with its original value:
http.request.uri.path ne raw.http.request.uri.path
This update provides immediate defense against a vulnerability affecting Citrix NetScaler ADC and Gateway appliances, deploying protection against improper input validation vectors.
Key Findings
CVE-2026-88771: An improper input validation vulnerability affecting Citrix NetScaler ADC and Gateway allows an unauthenticated attacker to execute arbitrary commands.
Impact
We strongly recommend that administrators apply the latest versions to fully secure origin servers. Additionally, customers should review configurations against applicable preconditions and follow standard incident response processes if signs of compromise are identified.
The Workers Web Crypto API now supports ML-KEM-768, ML-KEM-1024, ML-DSA-44, ML-DSA-65, and ML-DSA-87. ML-KEM establishes shared secrets, while ML-DSA signs and verifies data.
The opt-in API also adds key encapsulation and decapsulation methods, getPublicKey(), SubtleCrypto.supports(), and JSON Web Keys (JWKs) with the AKP key type.
Turn on the webcrypto_modern_algorithms compatibility flag to use these features:
Workers implements a subset of the evolving Modern Algorithms in the Web Cryptography API ↗︎ draft. ML-KEM-512 and the draft's other algorithms are not supported. The API may change as the draft evolves.
Containers now support the durable_object scheduling policy in public beta. This policy lets a Durable Object select the image and instance size for a Container at runtime instead of using one centrally managed configuration for the application.
To use custom images, configure the policy and one or more named images in Wrangler:
Wrangler prepares each image and exposes its immutable reference through ctx.container.images. Supply that reference and an instance size when you start the Container:
The durable_object policy also supports the new cloudflare/debian-trixie Cloudflare-managed image, which includes Node.js 24.20.0 on Debian Trixie slim. Start it directly without configuring a named image.
Durable Object-managed Container instances have independent lifecycles and do not participate in application-wide image rollouts.
For configuration, runtime sizing, snapshots, and update behavior, refer to Scheduling Policies.
Containers now support snapshot APIs in public beta for saving and restoring point-in-time filesystem state. Create a snapshot first, then pass it back to start() to restore files after container sleep, restart, or handoff to another Durable Object.
Use snapshotContainer() through the Durable Object Container API to capture the full container filesystem. Create a snapshot from a running Container, store its handle, and pass that handle to start() when you restore it later:
src/index.jsjs
import { DurableObject } from "cloudflare:workers";export class MyDurableObject extends DurableObject { async saveSnapshot() { // Create a snapshot from the running Container. const containerSnapshot = await this.ctx.container.snapshotContainer({}); await this.ctx.storage.put("containerSnapshot", containerSnapshot); } async restoreSnapshot() { // Restore the saved snapshot later. const containerSnapshot = await this.ctx.storage.get("containerSnapshot"); if (!containerSnapshot) { return; } this.ctx.container.start({ containerSnapshot, enableInternet: false }); }}
src/index.tsts
import { DurableObject } from "cloudflare:workers";export class MyDurableObject extends DurableObject { async saveSnapshot() { // Create a snapshot from the running Container. const containerSnapshot = await this.ctx.container.snapshotContainer({}); await this.ctx.storage.put("containerSnapshot", containerSnapshot); } async restoreSnapshot() { // Restore the saved snapshot later. const containerSnapshot = await this.ctx.storage.get<ContainerSnapshot>("containerSnapshot"); if (!containerSnapshot) { return; } this.ctx.container.start({ containerSnapshot, enableInternet: false }); }}
Snapshots are only supported for Container applications that use the durable_object scheduling policy. Snapshots are immutable, so create a new snapshot to persist filesystem changes made after a restore.
Sandbox SDK 1.0 is available. Your own Durable Object class now controls each sandbox container directly, through the Durable Object container API on this.ctx.container.
With 1.0, your class can:
Choose the image and instance size each time it starts a sandbox. One class can run sandboxes on different images, and a deploy does not restart sandboxes that are running.
Save the files of a sandbox as a snapshot, in public beta, and start the same sandbox or a new one from it.
Decide when each sandbox stops, for example when a task finishes, when its user goes idle, or after it saves a snapshot.
Run commands with streamed input and output, send them signals, and open terminals.
Serve previews from ports in the sandbox, with your own hostnames and authentication.
Handle outbound requests for each hostname in Worker code, so credentials and bindings stay in your Worker.
Expose only the methods that you want callers to use.
Run an agent in the same Durable Object, and give the model a tool that runs commands in the sandbox.
src/index.jsjs
import { Files } from "@cloudflare/sandbox";import { DurableObject } from "cloudflare:workers";export class MySandbox extends DurableObject { container; files; constructor(ctx, env) { super(ctx, env); if (!ctx.container) { throw new Error("No container is configured"); } this.container = ctx.container; this.files = new Files(ctx.container); } async run(script) { if (!this.container.running) { // Your code chooses the image, size, and network access. this.container.start({ image: this.container.images.sandbox, instance: "lite", enableInternet: false, }); } await this.files.writeFile("/tmp/task.sh", script); const proc = await this.container.exec(["sh", "/tmp/task.sh"]); return proc.output(); }}
src/index.tsts
import { Files } from "@cloudflare/sandbox";import { DurableObject } from "cloudflare:workers";export class MySandbox extends DurableObject<Env> { private readonly container: Container; private readonly files: Files; constructor(ctx: DurableObjectState, env: Env) { super(ctx, env); if (!ctx.container) { throw new Error("No container is configured"); } this.container = ctx.container; this.files = new Files(ctx.container); } async run(script: string) { if (!this.container.running) { // Your code chooses the image, size, and network access. this.container.start({ image: this.container.images.sandbox, instance: "lite", enableInternet: false, }); } await this.files.writeFile("/tmp/task.sh", script); const proc = await this.container.exec(["sh", "/tmp/task.sh"]); return proc.output(); }}
Files streams files in and out of the running sandbox.
S3Mount mounts an S3-compatible bucket, such as R2. Your Worker signs each storage request, so the credentials stay out of the sandbox.
DirectoryBackup saves a directory to R2 and restores it into any sandbox, including one on a newer image.
If you use Sandbox SDK 0.x
Your 0.x applications keep running, and @cloudflare/sandbox 0.x stays on npm. Sandbox SDK 0.x receives bug and security fixes until 2026-12-31, and its documentation stays at Sandbox SDK 0.x.
When you are ready, Migrate from Sandbox SDK 0.x shows the 1.0 code for each 0.x feature, including preview URLs, tunnels, background processes, terminals, backups, and the code interpreter. The guide keeps the preview URLs and named tunnels that your 0.x application created working. You can move every sandbox in one deploy, or run a 1.0 class next to your 0.x class and move sandboxes one at a time. The deploy that moves an existing class to the new policy is one-way, so the guide shows how to rehearse it first.
The Sandboxes documentation also covers Dynamic Workers, for untrusted code in JavaScript, Python, or WebAssembly.
This beta release includes the following changes and improvements:
Fixed an issue that could briefly block traffic to split tunnel excluded resources while the client was connecting or reconnecting.
Improved reauthentication reliability and fixed an issue where a reauthentication could force a new registration.
Improved client reaction to the current network lowering its MTU.
Added support for routing non-RFC 1918 local IPv4 networks through the WARP tunnel when unrestricted LAN inclusion is enabled by policy or MDM.
Improved DNS reliability on networks with lower MTUs by clamping the TCP maximum segment size (MSS) for DNS-over-HTTPS connections sent through the tunnel.
Improved API reliability by retrying requests dropped when reusing pooled connections.
The client no longer requires the Windows WLAN AutoConfig service to be running.
Implemented a service recovery mechanism backed by Windows scheduler task to start WARP service on system unlock if not already started.
Fixed slow captive portal checks causing the client service to become unresponsive or restart while connecting.
Fixed a race when switching tunnel protocols during key rotation that could prevent WireGuard from connecting.
Fixed the client continuing to report 'No network' after a successful manual disconnect.
Fixed Digital Experience Monitoring (DEX) HTTP tests failing TLS validation on Windows.
Fixed the client UI crashing at startup when it could not write to the Windows registry.
Fixed latency spikes and traffic interruptions during TPM-backed API authentication when hardware-backed registration is enabled.
Fixed trailing whitespace in BIOS serial numbers causing serial-number and client-certificate device posture checks to fail.
Fixed a client UI crash that could occur when the daemon connection was reset during an IPC request.
Fixed a startup crash when date formatting data for the system locale had not yet loaded.
