The defaults work for local development. Adjust Synapse in config/synapse.php when you need a different setup.
Dashboard URL-Change the default /synapse path and the middleware on its routes.
Agent Discovery-Choose directories to scan and agent classes to exclude. The defaults areapp/Ai/Agents and app/Agents.
Available Models-Add options to the playground's model selector.
Data Storage-Choose the database connection for Synapse's tables and the filesystem disk for attachments.
History Retention-Enable daily pruning and choose how long to keep conversations. Automatic pruning is off by default.
Access Control-Define who can use Synapse outside local through the viewSynapse gate in app/Providers/SynapseServiceProvider.php. Production is disabled by default.
Install Synapse and see every connection your Laravel AI agents make. Open source, built and maintained by Redberry, available under the MIT license




A development dashboard for agents built with the Laravel AI SDK. Chat with your agents in your browser, inspect tool calls, and track token usage using your app’s existing AI setup.
Building an agent means trying something, checking the result, and trying again. Without the right tool, every round takes longer than it should.
Every Small Change Means Another Terminal Run
Trying a new prompt or input often means running commands and sorting through output. Repeating that process slows down small changes.
It’s Hard to Follow Tool Calls
The final answer doesn’t show which tools the agent called, what it sent them, or what came back. You need those details to check its work.
Errors Don't Say What Went Wrong
A request or tool call fails, but the error alone doesn’t tell you what happened. Finding the cause means digging through logs.
Token Use Is Easy to Miss
Long prompts and large tool results use more tokens. Without counts for each reply and conversation, it’s hard to see what’s adding up.
.webp&w=1280&q=75&dpl=dpl_HpdNavjitZSXcCWHCC9srdUm9SFu)
Telescope and Horizon, but for agents. Created by Revaz Ghambarashvili, AI Lead at Redberry and creator of LarAgent, Synapse brings the agent development loop into your browser, using your existing Laravel AI SDK setup.
1
Iterate on Prompts
Change your agent’s instructions in code, refresh, and test again. Keep the conversation, response, and tool calls together while you work through the next revision.
2
Investigate Agent Behavior
Follow tool calls from their inputs to their results. See where a call failed, inspect the exception, and return to your code with the details you need.
3
Try Different Models
Choose another model for your next message without changing your agent’s code. Inspect its response, tool use, and token usage to judge how it handles the task.
4
Revisit and Share Test Runs
Reopen past conversations with their tool calls and errors intact. Share a link with teammates on the same development or staging instance to review what happened together.
Synapse reuses your providers and API keys, and ships with compiled frontend assets.
Add Synapse as a dev dependency. The installer publishes its configuration and access provider, then runs migrations. PHP 8.3+, Laravel 12.62+ or 13.15+, Laravel AI SDK ^1.3.
Your agents are listed at /synapse. Click one and start chatting.

Chat in the Browser
Send messages and watch supported responses stream into the conversation, with tool calls appearing as the agent works.
Switch Models Without Editing Code
Use the model selector to try another configured option for your next message. Each response records the model that actually ran.
Test with Attachments
Upload images, documents, or audio to test the inputs your agent and provider support.
Read Structured Responses
Inspect structured output as formatted, collapsible JSON.
Return to Past Conversations
Search messages and filter history by agent, status, tools, or date. Reopen the full thread, including recorded tool calls and errors.
Arguments
See the exact inputs the model passed to the tool.
Results
Inspect the returned data as formatted JSON.
Timing & Status
Check how long the call took and whether it completed or failed.
Errors
Read exception messages and expand the stack trace when a tool throws.
Tool Types
Distinguish your application's tools from provider tools such as web search.
Conversation Context
Follow calls where they happened, alongside the surrounding messages.
Open in Local Development
When Laravel's environment is set to local, the dashboard needs no login.
Control Shared Access
Outside local, the viewSynapse gate protects every dashboard and API route. Access is denied until you configure it.
Off in Production
Production routes stay disabled unless you set SYNAPSE_ENABLED=true. The access gate still applies when enabled.
Stored in Your Application
Playground history uses Synapse's own database tables. Uploaded files use your configured Laravel filesystem disk.
Shared History
Everyone with access to the same Synapse instance and database can see the same playground conversations.
Light, Dark, or System
Choose your theme or follow your operating system's appearance. Synapse remembers your preference.
Name Your Test Runs
Rename saved conversations so useful examples and failures are easier to find later.
Link to the Exact View
Send teammates a link to an agent's Prompt or Tools tab, or a filtered History view on the same instance.
Pick Up Where You Left Off
Refresh the page or return to an agent to reopen the conversation you were viewing in that browser.
Know When Memory Is Off
Agents without conversational memory are marked Stateless. Each message is sent without earlier conversation context.
Inspect your agent's setup and the details behind its responses, from tool definitions to token usage.
Automatic Agent Discovery
Find agent classes in your configured directories, listed with their provider, model, and tools. New classes appear when you refresh.
Resolved Prompts & Settings
Read the system instructions and configuration the SDK will use in the read-only info panel. Edit your application code, refresh, and see the changes.
Tool Definitions
Explore tool descriptions, parameters, and output schemas. Identify MCP tools, provider tools, and agents used as tools.
Available Reasoning
For supported models, open the Thinking pane to inspect reasoning exposed by the provider and retained with the conversation.
Token Usage & Timing
Check input and output tokens, conversation totals, and response time. Expand available cache and reasoning token breakdowns.
Automatic Agent Discovery
Find agent classes in your configured directories, listed with their provider, model, and tools. New classes appear when you refresh.
Resolved Prompts & Settings
Read the system instructions and configuration the SDK will use in the read-only info panel. Edit your application code, refresh, and see the changes.
Tool Definitions
Explore tool descriptions, parameters, and output schemas. Identify MCP tools, provider tools, and agents used as tools.
Available Reasoning
For supported models, open the Thinking pane to inspect reasoning exposed by the provider and retained with the conversation.