Weave Code
Filament Agents Laravel Package
packstub/filament-agents
A Filament plugin that adds AI agent management to your Laravel admin panel. Create and configure agents, define prompts, attach tools, run conversations, and monitor activity through native Filament resources and a clean, familiar interface built for Laravel.
An in-panel AI assistant and an MCP server for Filament v5 panels, built on laravel/ai and laravel/mcp through packstub/agents: one tool list for both, the panel's own authorization, approve-in-chat writes, live tables and charts in answers, per-workspace budgets and an operator limits page.
Frequently asked questions about Filament Agents
- What are the requirements to install packstub/filament-agents?
- The hard floors are Filament v5 and PHP 8.4 or higher, with no fallback path for older versions. Version 1.16 of the plugin requires packstub/agents ^1.9, so treat the plugin and engine as a pair and upgrade them together. Install via Composer and register the plugin in your Filament panel provider. Confirm both version requirements in every environment, including staging and CI, before committing to a plan.
- Does packstub/filament-agents work with Filament v4 or PHP 8.3?
- No. The package targets Filament v5 and PHP 8.4+ only, and there is no compatibility shim or downgrade path. If any environment in your fleet still runs Filament v4 or PHP 8.3, it must be upgraded first. Verify this before you start evaluating the plugin, because it is a gating constraint rather than a nice-to-have.
- What is the difference between packstub/filament-agents and packstub/agents?
- packstub/agents is the engine: the agent loop, tool contracts, budget enforcement, MCP server, migrations, and config. packstub/filament-agents is a thin Filament v5 adapter providing pages, Livewire components, resources, and navigation. Anything that must run outside a panel belongs in the engine, while your record shapes, filters, prompt content, and consent vocabulary belong in your own app behind AgentResource. Keeping that boundary clean is the main thing to enforce in code review.
- How does access control work for AI agents in Filament?
- The plugin integrates with Filament policies and standard Laravel authorization rather than standing up a parallel permission surface. The in-panel assistant inherits the panel's guard, tenancy, and policies via FilamentContext, and TenantSet still fires. This is a significant advantage for compliance and audit, because reviewers can reason about AI access using rules they already understand. Define authorization on your resources and the agent UI follows the same model.
- Does packstub/filament-agents support multi-tenancy?
- Yes. Because the assistant runs inside your Filament panel, it inherits Filament's tenancy behavior and the TenantSet event still fires. Existing tenant scoping on your resources and policies therefore applies to agent-driven actions. You should still test tenant switching explicitly, especially where tools mutate records, to confirm scope is enforced end to end.
- Can I expose my admin tools to AI agents over MCP?
- Yes. A single tool list feeds two channels: the in-panel assistant via laravel/ai and an MCP server via laravel/mcp. That means exposing your admin toolset over MCP is an incremental follow-on rather than a separate workstream. If you have no near-term MCP plans, though, recognize that you would still be maintaining that half of the package. Adopt it when both channels are on your roadmap.
- Which AI providers and models does the plugin support?
- Model support is constrained by the laravel/ai ecosystem, so you do not get provider-agnostic model integration out of this package. Agents store their model and settings in the resource, and you configure them per agent in the Filament form. Check that your preferred provider is available in laravel/ai before designing around a specific model. If multi-provider abstraction is a requirement, treat it as an application-level concern.
- How does tool approval work for mutating agent actions?
- As of v1.16, approval is fail-closed: a typed reply to a waiting proposal is evaluated by an application-owned rule, per-locale word lists that you can extend or replace, and an optional classifier. Ambiguous replies such as a hesitant yes or a qualified approval no longer approve the action. This is a behavior change on a minor upgrade, so operators trained on the old semantics need retraining. Localize the approval UX and test it in every locale you serve.
- Is packstub/filament-agents production ready?
- Treat it as an early-adopter bet rather than a mature dependency. It is MIT licensed with an active cadence but a single maintainer, modest star count, and effectively no dependents, so you are not following an established crowd. Before production, require a code review and establish a fork or vendoring escape hatch. Also confirm you have a named engineering owner for the domain layer behind AgentResource.
- Which databases are supported by packstub/filament-agents?
- SQLite, MySQL, and Postgres are supported, and CI runs all three. Postgres parity covers case-insensitive chat search, null-safe sorting of unclassified chats, and a guarded tenant lookup that avoids passing a slug into an integer key. Teams standardized on Postgres can adopt without worrying about subtly broken chat search on their fleet. Still run your own migration and query tests against the database you deploy.
Weaver
How can I help you explore Laravel packages today?