Loopers
Firewall for the Agentic Era
When software teams deploy autonomous agents, they run into a big problem: agents get stuck in loops, burn through API budgets, and execute unintended commands.
Today, teams rely on prompt instructions (which models ignore) or dashboards (which only show damage after it happens). Loopers is an independent firewall that sits between agents and LLMs to enforce hard spending and safety boundaries before actions occur.
Why Autonomous Agents Are Hard to Deploy Safely
Every engineering team wants agents that can write code, troubleshoot systems, or handle operations. But giving an LLM autonomy to run commands and make API calls creates unpredictable risks that current tools fail to solve.
Prompts Are Not Boundaries
Telling an agent "please do not delete files or exceed $10" in a system prompt does not work. When context gets long or the model gets confused, prompt rules are easily forgotten or bypassed.
Dashboards Alert Too Late
Observability tools and gateways record what happened and send a notification after an agent has burned hundreds of dollars in an infinite loop. They record the damage; they don't prevent it.
The Missing Guardrail
Teams need an active layer in the network path that holds requests, verifies budgets beforehand, and physically cuts connections the moment an agent starts repeating itself.
An Independent Firewall for AI Agents
In standard software, you don't trust user input—you put a firewall in front of your server. Loopers does the same for AI agents: an independent proxy sitting outside the agent that the model cannot bypass.
- Outside the Agent: Loopers Firewall runs independently of your application code. The agent cannot modify or bypass its own guardrails.
- Hard Spending Limits: Checks and reserves budget before any request is sent to the LLM provider. If a daily limit is set to $10, it physically cannot bill $10.01.
- Live Loop Detection: Watches streaming responses in real time. If an agent gets stuck repeating failed reasoning steps, Loopers disconnects the call immediately.
- Tool Access Control: Inspects tool calls (such as file edits, shell commands, or external APIs) and blocks unauthorized actions cleanly without crashing the agent.
Bottom-Up Developer Adoption · Team & Cloud Plans
We are building Loopers with a clear dual approach designed for natural developer adoption and enterprise monetization:
Loopers Open-Source Engine
Free, MIT-licensed standalone proxy written in Go. Individual developers use it to protect their personal API keys and local coding agents (Aider, Claude Code, Cursor) with zero code changes. This drives organic adoption and developer trust.
Loopers Teams & Cloud
Commercial control plane for engineering teams running multiple agents. Monetizes through shared team budgets, centralized audit logs, human approval workflows for sensitive actions, team access controls, and private on-prem/VPC deployments.
Agents Have Real Tools Now
In the last year, models transitioned from passive text chatbots to autonomous software operators with access to terminal shells, code editors, and production databases. As agents are given real authority to act, the risk shifts from minor typos to runaway costs and unintended system changes.
Raising Pre-Seed
Founded by Varad Khoriya and Mayank Agrawal. Mayank is an Application Lead at LTM with deep background in designing reliable, high-throughput backend infrastructure. We are actively raising our Pre-Seed round to expand our open-source engine, accelerate developer adoption, and ship our commercial control plane for engineering teams.
How Loopers Differs from Existing Tools
Understanding the difference between passive monitoring and active runtime prevention.
Prompt Instructions
e.g. AGENTS.md, system rules
- Suggestions, Not Walls: The model can ignore instructions when contexts expand.
- No Financial Control: Cannot track token spend or enforce hard monetary caps.
- Easily Bypassed: Vulnerable to jailbreaks and hallucinations.
Observability Dashboards
e.g. traditional AI gateways & loggers
- Passive Loggers: Show you charts of money after it is already spent.
- No Real-Time Severing: Cannot disconnect an active streaming loop mid-flight.
- Post-Mortem Only: Great for analytics, but cannot stop active incidents.
Loopers Firewall
Active Runtime Security
- Independent Network Proxy: Sits outside the agent where the model cannot touch it.
- Pre-Call Spending Checks: Verifies remaining balance before requests are forwarded.
- Mid-Stream Loop Breaker: Cuts the connection the moment a repetitive loop occurs.
- Tool Call Gating: Blocks unauthorized tool actions cleanly.
Ask Us for a 60-Second Live Demo
Since you just scanned our QR code, you're standing right in front of us! Look up and say hi to Varad or Mayank—we can run a live terminal demo right now showing an agent loop being severed in real time.
Good conversation starters right now:
- "Tell us about your Pre-Seed round, target terms, and current traction."
- "Can you show me a 60-second terminal demo of Loopers killing an agent loop?"
- "Why is an out-of-process proxy more reliable than an in-app software library?"