Claude Agent Laravel Laravel Package

Product Decisions This Supports

• Build vs. buy for conversational AI in a Laravel app. Adopting this package is really a decision to depend on Anthropic's Claude (via the CLI/API) as the model backend rather than a managed chat vendor — it's thin framework glue, not a platform commitment.
• Feature scoping for AI surfaces. It supports the "one chat turn + streaming" class of features: in-app assistants, drafting, summarization, support triage, internal copilots. It is not the layer that decides RAG, agents, or eval strategy.
• Deployment topology. The new bare option (CLAUDE_BARE) makes headless/containerized execution a first-class supported path instead of a local fork. That directly unlocks shipping AI features to queue workers, cron jobs, and Docker/Kubernetes workloads — an infrastructure decision, not just a config toggle.
• Secrets and key-management posture. exclude_env_keys turns "which env vars reach the model subprocess" into an explicit, reviewable choice rather than inherited SDK behavior (the SDK defaults to excluding ANTHROPIC_API_KEY). This is the knob legal/security will care about for per-tenant keys or secret isolation.
• Roadmap sequencing. It validates investing in Laravel-native AI plumbing now, keeping the vendor surface small, before committing to heavier orchestration later.
• Release-management policy. Pre-1.0 with a still-moving config schema → decide pinning strategy and whether the feature ships behind a flag or as GA.

When to Consider This Package

Adopt when:

• The product is Laravel/PHP and you want chat completions and streaming without onboarding an entire SDK ecosystem or a separate service.
• You're shipping an assistive/internal feature (copilot, summarizer, drafting) where time-to-market beats abstraction purity.
• Your runtime is containerized or ephemeral — bare mode in v0.2.1 targets exactly this.
• You need auditable control over which secrets reach the model process (exclude_env_keys).
• You value a dependency one Laravel developer can read end-to-end in an afternoon.

Look elsewhere when:

• You need multi-provider abstraction, model routing, or provider failover — use a dedicated AI gateway or the official provider SDK.
• You need RAG pipelines, agent tool-use loops, evals, or tracing/observability built in.
• You can't spawn subprocesses (strict serverless/FaaS) or need per-request guarantees a CLI wrapper can't provide.
• You need SLAs, a support contract, or a large hiring/support community.
• You're not on Laravel — the entire value is the framework glue.

Caveats: v0.2.x is pre-1.0; the config surface is still moving (two new options in a patch-level release). Pin to a minor and re-review on each bump. Changes here are config-only plus a lint fix, so 0.2.0 → 0.2.1 is low-risk.

How to Pitch It (Stakeholders)

Executives: "It's a small Laravel integration that lets us add Claude-powered features — drafting, summarization, an internal assistant — without a new platform commitment or vendor contract. This release makes it deployable in our containerized/headless environments and gives us explicit control over which API keys are exposed to the model process, which is the question legal and infra will ask. It's pre-1.0, so the plan is a flagged pilot on one surface before we widen it."

Engineering: "Thin Laravel wrapper over Claude chat with a CLI passthrough. v0.2.1 adds bare (CLAUDE_BARE) for headless/container runs — no more forking for CI or queue workers — and exclude_env_keys, so we stop relying on SDK default secret exclusion and state our own policy. Config-only change from 0.2.0 apart from a Pint fix in ClaudeChat. Still 0.x: pin the minor and expect the config schema to shift. Good fit for chat turn + streaming in-app; wrong tool if we need multi-provider routing, tool-use loops, or tracing."