# wirelatch: agent gateway, sandbox and audit trail > Agents, wired in. Every move on the record. wirelatch connects AI agents to your company's systems, gives each one a safe place to run, and records everything they do. Say yes to agents today, and to your auditors tomorrow. Status: early access. We are onboarding a small group of design partners. [Join the list](https://wirelatch.dev/#join). ## How it fits together Your agents (a coding agent fixing a bug, a support agent answering a customer, a data agent building a report, or any agent you bring) plug into wirelatch. wirelatch connects them to your code, tools and data, blocks anywhere that is not on your list, and writes every move to an append-only record that is ready for audit. Example record: - 09:41 Coding agent opened a pull request in payments-api - Allowed - 09:42 Data agent read the revenue dashboard - Allowed - 09:44 Support agent tried to reach an unknown site - Blocked - 09:45 Data agent asked to export customer data - Needs review ## The problem: agents are ready to work, the infrastructure around them isn't Putting an agent to work today means sharing keys, running it wherever there's room, letting it reach anything, and hoping someone can explain what it did. - **Keys everywhere.** Agents need access to do real work. Today that means passwords and tokens pasted into prompts, scripts and laptops. - **Nowhere safe to run.** Agents share machines with the things that matter most, one wrong command away from real trouble. - **No gate, no rules.** Nothing decides what an agent may reach. No limits, no reviews, and no off switch. - **No record, no audit.** When something goes wrong, nobody can say what happened. And when auditors ask, there's nothing to show them. ## The fix: one latch between your agents and your company Connect agents to the systems they need, give each one a safe place to work, and keep a record you can stand behind. ### Agent gateway: connect in minutes Plug agents into code, data and tools through one gateway. They reach exactly what you allow, and never see the keys. ### Agent sandbox: a safe place to run Deploy each agent into a sealed sandbox of its own. Pause it, bring it back, and it picks up right where it left off. ### Audit trail: every move on the record Everything is written down as it happens: what the agent did, what it touched, and what was stopped. ## See exactly what your agents are doing One live view of every agent across the company: what it's working on, what it reached, and what was stopped. Filter by agent, see what was blocked or is waiting for review, and export the record for auditors. The record is append-only, so nothing gets lost. ## AI is coming to every audit. Be ready. Soon regulators, customers and your own auditors will ask what your AI did, and why it was allowed. With wirelatch, the answer is already written down. - **The whole story.** What every agent did, where, and when. All in one place. - **Can't be rewritten.** The record is append-only. What happened stays on the record. - **Rules you can show.** What each agent may reach is written in plain language, ready to review. - **Evidence on request.** Hand auditors a clear record of what every agent did, and why it was allowed. ## How it works: plug in, set the rules, watch it work 1. **Connect.** Link your code, data and tools once. Keys stay with wirelatch. Agents get access through the gateway, never the credentials. 2. **Set the rules.** Choose what each agent may reach, in plain language. Send anything sensitive for a review. Anything you didn't allow stays shut. 3. **Watch it work.** Deploy agents and see every move. Each one runs in its own sandbox, with a record your security team and auditors can trust. ## Works with Agents: Claude Code, Codex, Gemini CLI, agents built on MCP, and your own agents. Systems: GitHub, Slack, Linear, Jira, Postgres, MCP servers and your own APIs. ## Common questions ### What is wirelatch? wirelatch is an agent gateway, sandbox and audit trail in one. It connects AI agents to your company's code, tools and data, runs each agent in its own isolated sandbox, and keeps an append-only record of everything they do. ### How do I give AI agents access to company systems without sharing keys? Agents reach your systems through the wirelatch gateway. Passwords, tokens and API keys stay in wirelatch. Each agent gets access to exactly what you allow, and never sees the secrets. ### What is an agent gateway, and does it work with MCP? An agent gateway sits between AI agents and the tools they use, like MCP servers, APIs and databases. It decides what each agent may reach, blocks the rest and logs every call. wirelatch works with MCP, so it can be your MCP gateway too. ### How do you sandbox AI agents? Every agent runs in its own isolated sandbox runtime, away from your laptops and production servers. Pause it, bring it back, and it picks up right where it left off. ### How do I deploy AI agents safely? Bring the agent, connect the systems it needs, set the rules and deploy. It runs in its own sandbox, reaches only what you allowed, and every move lands on the record. ### How do I audit what AI agents did? Every action is written down as it happens: what the agent did, which system it touched, and whether it was allowed, blocked or sent for review. Hand the record to auditors, customers or your security team. ### Which agents and tools does wirelatch work with? Coding agents like Claude Code, Codex and Gemini CLI, agents built on MCP, and your own agents. On the other side: GitHub, Slack, Linear, Jira, Postgres, MCP servers and your own APIs. ### Can I use wirelatch today? wirelatch is in early access, and we're onboarding a small group of design partners now. Join the list and we'll reach out. ## Early access Be first to wire in your agents. Work directly with the founders, shape what we build next, and get in before everyone else. [Join early access](https://wirelatch.dev/#join).