Laravel AI Agents: Streaming, Token Budgets & Typed Output | Mohamed Said [Skip to content](#main) [ ![](https://cdn.msaied.com/01KT78WE565VEMM3PSNQAAB0MH.png) Mohamed SaidLaravel Backend Engineer ](https://www.msaied.com/public) - [Home](https://www.msaied.com/public) - [Projects](https://www.msaied.com/public/projects) - [Articles](https://www.msaied.com/public/articles) - [Certificates](https://www.msaied.com/public/certificates) - [About](https://www.msaied.com/public#about) [ Contact](https://www.msaied.com/public#contact) Menu Menu ---- Close - [HomeStart here](https://www.msaied.com/public) - [ProjectsCase studies](https://www.msaied.com/public/projects) - [ArticlesEngineering notes](https://www.msaied.com/public/articles) - [CertificatesCredentials](https://www.msaied.com/public/certificates) - [AboutHow I work](https://www.msaied.com/public#about) - [ContactGet in touch](https://www.msaied.com/public#contact) [Start a conversation](https://www.msaied.com/public#contact) [WhatsApp](https://wa.me/201094619204) [Email](mailto:hello@msaied.com) 1. [Home](https://www.msaied.com/public) 2. / 3. [Articles](https://www.msaied.com/public/articles) 4. / 5. Production AI Agents in Laravel: Streaming, Token Budgets, and Structured Output Contracts Production AI Agents in Laravel: Streaming, Token Budgets, and Structured Output Contracts =========================================================================================== Ship reliable AI agents in Laravel by combining streamed responses, hard token budgets, and typed structured-output contracts — without letting LLM non-determinism bleed into your domain layer. ![](https://cdn.msaied.com/01M22N44A70A5MC2S599JP0MPH.webp) [Mohamed Said](https://www.msaied.com/public#person) Published 22 Jun 2026 · Updated 22 Jun 2026 · 3 min read ShareCopy linkCopied ![Production AI Agents in Laravel: Streaming, Token Budgets, and Structured Output Contracts](https://cdn.msaied.com/262/4220ef4ee3875717c6384f298d891a3f.png) On this page +1. [The Problem With Naive AI Integration](#the-problem-with-naive-ai-integration) 2. [Streaming Responses to the Browser](#streaming-responses-to-the-browser) 3. [Token Budget Guard](#token-budget-guard) 4. [Structured Output Contracts](#structured-output-contracts) 5. [Keeping the Domain Clean](#keeping-the-domain-clean) 6. [Takeaways](#takeaways) The Problem With Naive AI Integration ------------------------------------- Most Laravel + LLM tutorials stop at `Http::post('https://api.openai.com/v1/chat/completions', [...])` and call it done. That works for demos. In production you need three things the tutorials skip: 1. **Streaming** — users shouldn't stare at a spinner for 8 seconds. 2. **Token budgets** — unbounded prompts destroy your billing and latency SLAs. 3. **Structured output contracts** — raw JSON strings from an LLM are not domain objects. Let's solve all three without a heavy third-party SDK. --- Streaming Responses to the Browser ---------------------------------- OpenAI's `stream: true` returns server-sent events. Laravel's `StreamedResponse` pipes them straight to the client. ```php // app/Http/Controllers/AgentController.php public function stream(Request $request): StreamedResponse { $prompt = $request->validated()['prompt']; return response()->stream(function () use ($prompt) { $stream = Http::withToken(config('services.openai.key')) ->withOptions(['stream' => true]) ->post('https://api.openai.com/v1/chat/completions', [ 'model' => 'gpt-4o-mini', 'stream' => true, 'messages' => [['role' => 'user', 'content' => $prompt]], ])->toPsrResponse()->getBody(); while (! $stream->eof()) { $line = trim($stream->read(512)); if (str_starts_with($line, 'data: ')) { $payload = substr($line, 6); if ($payload === '[DONE]') break; $delta = json_decode($payload, true)['choices'][0]['delta']['content'] ?? ''; echo "data: {$delta}\n\n"; ob_flush(); flush(); } } }, 200, ['Content-Type' => 'text/event-stream', 'X-Accel-Buffering' => 'no']); } ``` The `X-Accel-Buffering: no` header is essential when Nginx sits in front — without it, Nginx buffers the whole response. --- Token Budget Guard ------------------ Never let user input dictate prompt size. Enforce a budget before the HTTP call. ```php // app/AI/TokenBudget.php final class TokenBudget { private const CHARS_PER_TOKEN = 4; // rough heuristic private const MAX_INPUT_TOKENS = 1_500; public static function enforce(string $text): string { $limit = self::MAX_INPUT_TOKENS * self::CHARS_PER_TOKEN; if (strlen($text) post('https://api.openai.com/v1/chat/completions', [ 'model' => 'gpt-4o-mini', 'max_tokens' => 256, 'response_format' => [ 'type' => 'json_schema', 'json_schema' => [ 'name' => 'sentiment_result', 'strict' => true, 'schema' => [ 'type' => 'object', 'properties' => [ 'label' => ['type' => 'string', 'enum' => ['positive','neutral','negative']], 'score' => ['type' => 'number'], 'reason' => ['type' => 'string'], ], 'required' => ['label','score','reason'], 'additionalProperties'=> false, ], ], ], 'messages' => [ ['role' => 'system', 'content' => 'Analyse sentiment. Reply only with the JSON schema.'], ['role' => 'user', 'content' => $text], ], ])->throw()->json(); $raw = json_decode( $response['choices'][0]['message']['content'], true, flags: JSON_THROW_ON_ERROR ); return SentimentResult::fromArray($raw); } } ``` With `strict: true` the model is constrained to the schema at the API level — you still validate on your side, but you'll rarely see a mismatch. --- Keeping the Domain Clean ------------------------ The `SentimentAgent` returns a typed DTO. Nothing in your domain layer touches raw LLM strings. If you swap providers tomorrow, only the agent changes — every caller keeps working. Wrap the agent in a queued job for non-interactive workloads, and inject it via the service container so Pest can swap in a fake: ```php // tests/Feature/SentimentTest.php it('classifies positive text', function () { $this->instance(SentimentAgent::class, new class { public function analyse(string $text): SentimentResult { return new SentimentResult('positive', 0.95, 'stub'); } }); $result = app(SentimentAgent::class)->analyse('Great product!'); expect($result->label)->toBe('positive'); }); ``` --- Takeaways --------- - Stream via `response()->stream()` and set `X-Accel-Buffering: no` for Nginx. - Enforce token budgets *before* the HTTP call, not after. - Use `response_format.json_schema` with `strict: true` to pin model output shape. - Map LLM JSON to typed readonly DTOs immediately — keep raw strings out of your domain. - Inject agents through the container so tests can swap fakes without HTTP calls. - [laravel](https://www.msaied.com/public/articles?search=laravel) - [ai](https://www.msaied.com/public/articles?search=ai) - [llm](https://www.msaied.com/public/articles?search=llm) - [php](https://www.msaied.com/public/articles?search=php) Frequently asked questions --------------------------- Does `response\_format: json\_schema` work with all OpenAI models?No. As of the current API, structured output with `strict: true` is supported on `gpt-4o`, `gpt-4o-mini`, and later snapshots. Older models like `gpt-3.5-turbo` support `json\_object` mode only, which does not enforce a schema. How do I handle streaming in a queued job rather than an HTTP response?In a job you don't need SSE. Disable streaming (`stream: false`), collect the full completion, then persist or broadcast the result. Streaming is only valuable when a human is waiting in real time. Is the 4-characters-per-token heuristic accurate enough for production?It's a conservative approximation for English text. For precise budgeting, use a tokeniser library such as `yethee/tiktoken` which implements the actual BPE encoding. The heuristic is fine as a cheap pre-flight guard. ![Mohamed Said](https://cdn.msaied.com/01M22N44A70A5MC2S599JP0MPH.webp)About the author ---------------- [Mohamed Said](https://www.msaied.com/public#person)Senior Backend Engineer specializing in Laravel, scalable SaaS platforms, APIs, and cloud infrastructure. I build secure, high-performance web applications that help businesses grow. [About](https://www.msaied.com/public#about) [GitHub ↗](https://github.com/EG-Mohamed) [LinkedIn ↗](https://www.linkedin.com/in/msaiedm/) [WhatsApp ↗](https://wa.me/201094619204) [Email Address ↗](mailto:hello@msaied.com) [My CV ↗](https://drive.google.com/file/u/0/d/1MF20IPRJyzfy32mhEutjL5EpSls0w2Q8/view) [Previous articleFilament v3 to v4: Breaking Changes and Practical Refactor Patterns](https://www.msaied.com/public/articles/filament-v3-to-v4-breaking-changes-and-practical-refactor-patterns) [Next articleMulti-Tenant SaaS in Laravel: Isolating Tenant State with Scoped Singletons](https://www.msaied.com/public/articles/multi-tenant-saas-in-laravel-isolating-tenant-state-with-scoped-singletons) On this page ------------- 1. [The Problem With Naive AI Integration](#the-problem-with-naive-ai-integration) 2. [Streaming Responses to the Browser](#streaming-responses-to-the-browser) 3. [Token Budget Guard](#token-budget-guard) 4. [Structured Output Contracts](#structured-output-contracts) 5. [Keeping the Domain Clean](#keeping-the-domain-clean) 6. [Takeaways](#takeaways) ### Have a technical challenge? Tell me what you’re building. I reply within two working days. [Start a conversation](https://www.msaied.com/public#contact) Related articles ----------------- [ ![](https://cdn.msaied.com/740/cce86edc21eddcbdd2f2454fadaf9c70.png) · 3 min read### The Pipeline Pattern in Laravel: Custom Pipelines Beyond Middleware 5 Oct 2026 ](https://www.msaied.com/public/articles/the-pipeline-pattern-in-laravel-custom-pipelines-beyond-middleware-1) [ ![](https://cdn.msaied.com/739/2d6897fdcdcf090613f96f72a64b8a78.png) · 4 min read### MySQL Full-Text Search in Laravel: Indexes, Relevance Scoring, and Boolean Mode 4 Oct 2026 ](https://www.msaied.com/public/articles/mysql-full-text-search-in-laravel-indexes-relevance-scoring-and-boolean-mode) [ ![](https://cdn.msaied.com/738/073696a3fefe18bec825beec5ac658f5.png) · 4 min read### Laravel Queue Rate-Limited Middleware: Throttling Jobs Without Losing Work 4 Oct 2026 ](https://www.msaied.com/public/articles/laravel-queue-rate-limited-middleware-throttling-jobs-without-losing-work) Have a technical challenge? ---------------------------- Tell me what you’re building. I reply within two working days. [Discuss your project ↗](https://www.msaied.com/public#contact) © 2026 Mohamed Said · Built with Laravel, meant to last.Senior Backend Engineer specializing in Laravel, scalable SaaS platforms, APIs, and cloud infrastructure. I build secure, high-performance web applications that help businesses grow. - [Home](https://www.msaied.com/public) - [Articles](https://www.msaied.com/public/articles) - [Certificates](https://www.msaied.com/public/certificates) - [GitHub](https://github.com/EG-Mohamed) - [LinkedIn](https://www.linkedin.com/in/msaiedm/) - [WhatsApp](https://wa.me/201094619204) - [Email Address](mailto:hello@msaied.com) - [My CV](https://drive.google.com/file/u/0/d/1MF20IPRJyzfy32mhEutjL5EpSls0w2Q8/view) - [Sitemap](https://www.msaied.com/public/sitemap.xml)