Welcome to TameFlare

A transparent proxy firewall for AI agents. Intercept all outbound HTTP traffic, enforce per-agent permissions, require human approval for risky actions, and keep an audit trail of everything your agents do. Zero code changes required.

What is TameFlare?

AI agents are increasingly autonomous. They can create pull requests, send emails, transfer money, and provision infrastructure — all without a human in the loop. That's powerful, but it's also risky.

TameFlare sits between your agents and the tools they use. Every action an agent wants to take is sent to TameFlare first. TameFlare evaluates the action against your policies and returns one of three decisions:

Policies are configured in the dashboard via a 4-step builder and stored as JSON. They can match on action type, target environment, parameters, risk level, time of day, and more. TameFlare uses a deny-wins model: if any policy says deny, the action is denied regardless of other policies.

How it works

# 1. Sign up at tameflare.com/register

# 2. Create a gateway in the dashboard (Gateways → Create gateway)

# 3. Run your agent through the proxy
npx @tameflare/cli run --name "DevOps Bot" python my_agent.py

# All outbound HTTP is now proxied:
# domain → connector → parse action → check permission
# allowed:          inject credentials → forward
# denied:           return 403
# require_approval: hold connection → wait for human

The gateway is a Go binary that acts as a transparent HTTP/HTTPS proxy. It intercepts all outbound traffic from your agent, parses requests via connectors (GitHub, OpenAI, Anthropic, Stripe, Slack, or any HTTP API), checks per-agent permissions, and injects credentials from an encrypted vault. The agent never sees real API keys.

The control plane is a Next.js application that hosts the dashboard, v1 SDK API, policy engine, and audit trail. For v2 proxy mode, the gateway handles enforcement directly. Both modes can coexist.

What can you control?

TameFlare's policy engine is generic. It evaluates fields like action_spec.type and parameters.* against conditions you define. This means you can write policies for any domain:

See the Writing Policies page for real-world examples across all these domains.

Key capabilities

Documentation