Security by Isolation

Tamer.ai is built on a "Local-First" philosophy. We believe that control over your AI agents should never come at the cost of exposing your intellectual property or sensitive credentials.

1. 1. Air-Gapped Secrets

   Your API keys (Anthropic, OpenAI, Google) never leave your machine. Tamer wraps your agent locally, meaning the agent communicates directly with the LLM providers from your terminal. The Tamer server only sees redacted metadata and the PTY stream.

2. 2. Zero-Trust Orchestration

   When using our Multi-Agent Pipeline, Tamer acts as a secure air-lock. If you have Claude coding and Gemini testing, they operate in separate isolated workers. The Master Agent synchronizes them without sharing unredacted memory between models.

3. 3. Path Protection & Sandboxing

   Tamer implements granular file-system rules. You can define "No-Go" zones (like .ssh/ or .env) that agents are physically blocked from reading or modifying, regardless of their own internal logic.

4. 4. End-to-End Control

   Every dangerous action (bash command, file deletion) requires a Human-In-The-Loop approval. Your phone acts as a physical security key; no agent can execute code without your explicit "Hold to Approve" confirmation.

5. 5. Kernel-Level Sandbox

   Tamer confines every agent process inside a kernel-enforced sandbox. The agent physically cannot reach files or syscalls outside the allowed perimeter.

   Linux

   • Landlock LSM — filesystem access control
   • seccomp-bpf — syscall filtering
   • bubblewrap — optional user-namespace sandbox

   Windows

   • Job Object — resource & process limits
   • Low Integrity Process — reduced privilege token
   • AppContainer — network & filesystem isolation