Understand
See the request, prepared model context, memory and skills in every turn.
Local-first AI agent security for macOS
See what your AI agents are doing.
Know whether it is safe.
AgentReins turns prompts, model routes, memory access, tool calls, processes, network traffic, and code changes into one evidence-backed live task.
●Download for Apple SiliconM1, M2, M3 and M4 · macOS 13+
●Download for Intel MacIntel processor · macOS 13+
Version 1.1.0 · Developer ID signed · Apple notarized
● Runs locally● Open source● Evidence, not guesses
LIVE OPERATIONS CONSOLE
3 agents observed · 1 task active
LIVE TASK · EVIDENCE CHAIN
01
User requestCreate and compile a C program
02
Context preparedMemory, skills and 14 project files included
03
Model connectionPrivate relay → claimed DeepSeek model
04
Shell executionclang hello.c -o hello
05
Independent verificationWaiting for build result
ONE TASK. THE COMPLETE STORY.
From human intent to machine impact.
Follow the causal chain instead of searching disconnected logs.
Trace
Connect tools, MCP servers, processes, network destinations and file changes.
Verify
Separate “the agent says done” from results independently proven to work.
MODEL ROUTE SECURITY
Know exactly what left your machine.
AgentReins identifies the gateway, claimed upstream model, context categories, token volume, and sensitive values found in captured content.
Your agent68,970 input tokens
→Private relayOperator unverified
→Claimed modelIdentity unproven
Make every AI agent understandable,
verifiable, and trustworthy.
Built in public. Local-first by design.
Follow AgentReins on GitHub ↗