Laravel AI SDK adds human-in-the-loop tool approval. At Laracon US 2026 in Boston, the Laravel team announced a human-in-the-loop (HITL) API for the Laravel AI SDK. Agents can now pause before executing a tool and wait for a person to approve it, reject it, or edit its arguments. The feature landed in laravel/ai#773 and shipped in v0.10.0 on July 21. Before this, once an agent started calling tools, it ran to completion on its own. That's fine when the tools only read data. It's a harder sell for a tool that deletes a file, refunds a payment, or emails a customer. From the Laravel blog: Until now, an AI agent built with the Laravel AI SDK ran on autopilot: once it started executing tools, there was no way to step in. The new human-in-the-loop (HITL) API lets you intercept specific agent actions and require a decision before the agent proceeds, so you can approve, deny, or modify what happens next instead of finding out after the fact. #Marking a tool as approvable. Approval is opt-in per tool. Implement the Approvable contract, use the InteractsWithApprovals trait, and the tool requires approval by default: use Laravel\Ai\Concerns\InteractsWithApprovals; use Laravel\Ai\Contracts\Approvable; use Laravel\Ai\Contracts\Tool; class IssueRefund implements Approvable, Tool { use InteractsWithApprovals; public function handle(Request $request): Stringable|string { $order = Order::findOrFail($request['order_id']); $order->refund($request['amount']); return "Refunded {$request['amount']} on order {$order->id}."; } // ... } Asking every time is often too blunt. To decide based on the arguments, define a needsApproval method that returns a boolean or an Approval instance carrying a reason: use Laravel\Ai\Approvals\Approval; protected function needsApproval(Request $request): Approval|bool { return $request['amount'] <= 2000 ? false : Approval::required('Refunds over $20 need a manager.'); } You can also override the requirement where the agent registers its tools, using withoutApproval and requireApproval: public function tools: iterable { return [ (new LookUpOrder)->withoutApproval, (new IssueRefund)->requireApproval('Every refund gets reviewed.'), ]; } #Approving, rejecting, or editing a call. When the model calls an approvable tool, the agent stops short of running it and hands the pending calls back on the response. Each one carries the tool call ID, the tool name, the arguments, and the reason: $response = (new SupportAgent) ->forUser($user) ->prompt('Refund the damaged headphones on order 4192.'); if ($response->hasPendingApprovals { foreach ($response->pendingApprovals as $approval) { // $approval->id // $approval->tool // $approval->arguments // $approval->reason } } To resume, continue the conversation and pass a Decisions instance keyed by tool call ID: use Laravel\Ai\Approvals\Decision; use Laravel\Ai\Approvals\Decisions; $response = (new SupportAgent) ->continue($conversationId, as: $user) ->prompt(Decisions::from([ 'call_abc' => Decision::approve, 'call_ghi' => Decision::reject('This order is outside the return window.'), ])); The booleans true and false work as shorthand for approve and reject. Every pending call requires a decision, or you get an ApprovalMismatchException, which also covers unknown IDs and calls that have already been resolved. If you don't want to enumerate all of them, approveRemaining and rejectRemaining set a default for the rest: $decisions = Decisions::from([ 'call_abc' => true, ])->rejectRemaining('Not approved.'); A rejection that includes a result string goes back to the model so it can keep responding. A rejection with no result records the rejection and stops the generation loop there. #Details worth knowing. * The agent has to be Conversational with its history persisted; otherwise, there is nothing to resume the paused call from. The RemembersConversations trait handles that. * Approval is supported by prompt, stream, queue, broadcast, broadcastNow, and broadcastOnQueue. * During streaming and broadcasting, a pause arrives as a tool_approval_request event. With the Vercel AI SDK stream protocol, approvals are emitted using that protocol's native tool approval parts. * Queued agents pass the response to the then callback and dispatch a ToolApprovalRequested event. * Pauses are per call, not per step. Tools in the same step that don't require approval run immediately, so keep external side effects idempotent using $request->toolCallId. * Laravel records an approved tool's result before it asks the model to continue. If generation fails after that point, the approval is already spent, so resume with an ordinary text prompt instead of resubmitting the decisions. Upgrading to 0.10 brings a couple of breaking changes: a new nullable approval_state column on the conversation messages table and a storeApprovalResults method that custom ConversationStore implementations must provide. If you're adding an approval step to an agent you already have, the documentation includes a full flow: a pair of routes that accept either a new message or a set of decisions from a chat screen. The rest of the Laracon US announcements are on the Laravel blog, and the code lives in the laravel/ai repository. Staff Writer at Laravel News and Full stack web developer.

Announcing Laravel Boost. Laravel LLC. is releasing the public beta of Laravel Boost two weeks after Laravel LLC. previewed it at Laracon US 2025. Boost runs in your project, provides a Laravel-aware model context protocol (MCP) server, and gives AI agents a set of tools, versioned guidelines, and version-specific documentation so they behave like an experienced Laravel developer instead of a search engine. By adding this package to your application, your AI coding agent can: run Tinker, query the database, inspect schema, search versioned Laravel docs, read browser logs, and generate tests and code that follow your project conventions so the output is helpful and reviewable. Quick facts about boost. * Public beta is available now as a composer package. * Boost is supported in Laravel 10, 11, and 12 on PHP 8.1 or higher. * It ships: * A Laravel-specific MCP server with 15+ tools that let AI agents inspect and act inside your app. * Laravel-maintained AI guidelines, specific to your installed packages and versions. * Access to over 17,000 pieces of vectorized Laravel ecosystem documentation, specific to your installed packages and versions. * Boost is free and open source: Send Laravel LLC. your issues, PRs, and honest feedback (here's its contribution guide). Getting started. composer require laravel/boost -dev php artisan boost:install The installer is interactive and will auto-detect IDEs and agents already in the repo and let you opt in to the pieces you want. Laravel LLC. do not force opinionated style rules on existing projects by default. Review what it installs, then enable the bits that make sense for you and your team. Boost features in detail. Laravel-Specific MCP. Boost's Laravel-specific MCP exposes 15+ tools to the agent, including application info, search docs, Tinker, browser logs, database queries, database schema, list Artisan commands, last errors, list routes, read configuration values, read log entries, and report feedback. These tools let the agent reason with the real state of your app. Documentation API. Boost gives you a documentation API built from Laravel ecosystem documentation. This documentation API has more than 17,000 version-specific data entries to help you get higher quality code. Laravel LLC. ingest, chunk, and vectorize docs so the agent can return guidance targeted to the exact major versions your project uses. Laravel-Maintained AI guidelines. These are composable, version-specific rules that nudge the agent to follow conventions, add tests when appropriate, use the correct APIs, and avoid common AI pitfalls. Fixing the AI context gap. Large language models are surprisingly good at PHP because there is a lot of PHP training data. That helps, but it's not enough. Without context, they invent APIs, misuse framework idioms, or skip tests. Boost fixes the context problem: It gives the agent the exact docs and programmatic access to your app, so the output is much more accurate and actionable. In testing, Laravel LLC. saw agents use search docs to find the correct API for the exact package version, inspect the database schema, run Tinker to create fixtures, and read storage/logs/browser.log to triage frontend errors. The result is fewer hallucinations and less wasted time cleaning up AI-made changes. Real capabilities you can expect. Beyond the AI hype, Laravel LLC. built Boost to help you achieve tangible outcomes: * Generate code that uses the right framework API for your exact Inertia or Livewire version. * Ask the agent to fix a white screen of death. The agent can fetch the last Laravel error, read recent browser logs, locate a typo, and propose a fix. * Ask for a rate limit and get an implementation that matches your Laravel version, plus tests when appropriate. * Use Tinker to create fixtures or inspect models as part of a conversational session. These are repeatable capabilities Laravel LLC. validated during testing. The agent will still be wrong sometimes, but Boost reduces the errors and nudges the agent to do the right things. Beta caveats (read this before you flip it on). Boost is in public beta, so you can expect some rough edges and behavioral changes as Laravel LLC. gather feedback and ship updates. The documentation ingest currently works at the major-version level only, so if your code depends on minor or patch differences, you will need to double-check that any generated code is correct. And, no matter how good the output looks, treat all generated code as a draft: run your tests, review the diffs, and make code review a non-negotiable part of the process. Help Laravel LLC. shape boost. Boost is pragmatic, but it's not a replacement for engineering judgment. It reduces busywork, provides high-context suggestions, and helps you stay in flow. It will not be perfect (hence the public beta), but you will learn faster whether it helps by trying it. Install it, poke it, break it, and tell Laravel LLC. what Laravel LLC. need to fix. Boost includes a "report feedback" tool for frictionless feedback that goes straight from your editor to the Laravel team. Frustrated by Boost or AI? Tell your agent: "Give Boost feedback: browser logs search failed when I tried to debug an Inertia form error." Or, our favorite: "Give Boost feedback: this is amazing!" Tell Laravel LLC. where the agent failed (or nailed it). Show Laravel LLC. the input, the tools used, and the (in)correct output when you can. If Boost saves you an afternoon of grunt work, Laravel LLC. will take that as a win. If it trips you up, tell Laravel LLC. where so Laravel LLC. can make it better. AI coding agents can now pass every Laravel Boost eval. Next up: measuring idiomatic Laravel and correct code per token. Pushpak Chhajed Connect any Laravel AI agent to an MCP server using the Laravel AI SDK and MCP. Supports bearer and OAuth auth, STDIO and HTTP transports, and built-in caching. Pushpak Chhajed The Laravel AI SDK is the LangChain PHP alternative: agents, tools, memory, streaming, and multi-agent workflows, entirely in PHP. No Python required. Ana Tavares

Laravel 13.20: native image processing and the new Image facade. Updated: July 24, 2026 Laravel 13.20 introduces a first-party Image facade, providing native image resizing, conversion, and storage. This update dramatically streamlines media management and reduces reliance on third-party libraries. 1. Introduction: A new era for media management in Laravel. Image manipulation has long been a foundational, yet often complex, aspect of modern web development. Developers frequently grapple with resizing, optimizing, and converting images for various platforms and user experiences, typically relying on a patchwork of external libraries and services. This common challenge often introduces extra dependencies and configuration overhead, complicating project setup and maintenance. Laravel 13.20 arrives as a significant release, directly addressing these long-standing pain points with the introduction of its groundbreaking first-party Image facade. This update marks a pivotal moment, fundamentally shifting how developers manage media within the framework. The core promise of Laravel 13.20 is a dramatically simplified approach to image processing. By integrating these capabilities natively, the framework significantly reduces the need for external dependencies, thereby streamlining media management directly within the application's ecosystem. This powerful new Image facade enables developers to effortlessly perform essential operations such as resizing, converting, and storing images, all without leaving the familiar Laravel environment. 2. Deep dive: understanding the native image facade. Laravel's new Image facade stands as the framework's official, first-party solution for robust image processing. It's designed to provide a comprehensive, yet intuitive, API for common image manipulation tasks, moving these operations from a third-party concern to a core framework feature. This integration means developers can expect consistent behavior and deep alignment with Laravel's conventions. The facade's core capabilities are meticulously designed to cover the most frequent image processing needs: * Image Resizing: Developers can precisely define image dimensions using resize(width, height), ensuring images fit specific layouts. Beyond fixed dimensions, the facade supports intelligent resizing methods like fit(width, height) to fill given dimensions while maintaining aspect ratio, or crop(width, height, x, y) for more granular control. It also allows for fluid resizing based on a single dimension, such as width(800) or height(600), with the other dimension automatically adjusted to maintain proportionality. * Image Conversion: The Image facade offers seamless conversion between various popular image formats, including JPEG, PNG, and the increasingly important WebP. This functionality is critical for optimizing web performance and ensuring cross-browser compatibility. Furthermore, developers can specify quality settings during conversion (e.g., toWebp(75)) to strike the perfect balance between file size and visual fidelity, crucial for efficient content delivery. * Direct Storage Integration: A significant advantage of this native facade is its seamless integration with Laravel's existing Storage facade. Processed images can be directly stored to any configured filesystem disk, whether it's a local disk, Amazon S3, DigitalOcean Spaces, or any other compatible driver. This eliminates the need for manual file handling after processing, streamlining the entire workflow from upload to persistent storage. Under the hood, this