# AgentGarm > Runtime governance, host sentinel and cost attribution for AI agents. Free to use; prompts and evidence stay on the user's machines. ## Pages - [Runtime governance for AI agents](https://agentgarm.com/): AgentGarm decides and records what AI coding agents do on your machines: policy, approvals and cost, enforced where they run. - [How it works](https://agentgarm.com/how-it-works): The layers AgentGarm governs an agent through, how it correlates one task, and where each fail mode is enforced. - [Agent coverage](https://agentgarm.com/coverage): Each agent AgentGarm has a hook adapter for, the level it is rated at today, and its gaps. - [Cost attribution](https://agentgarm.com/finops): What AI agents cost, per task and per developer, with budgets and FOCUS export. - [Editions](https://agentgarm.com/editions): AgentGarm Community, Central Lite and Central: what each includes and how each is offered. - [Changelog](https://agentgarm.com/changelog): Each AgentGarm release, newest first, with its notes and its date, from the first public release, Community 0.1.0. - [Join the beta](https://agentgarm.com/beta): Try AgentGarm before its first release: who the beta is for, what it tests, and one prefilled email to join it. - [For AI agents](https://agentgarm.com/agents): Where an AI agent, or the person pointing one here, finds this site's llms.txt, its full text, the policy schema and the release feed. - [About](https://agentgarm.com/about): Who makes AgentGarm, how to reach us, and where the name comes from. - [Download](https://agentgarm.com/download): AgentGarm for macOS, Windows and Linux, and Central Lite, with checksums and signatures. - [Verify a download](https://agentgarm.com/download/verify): How to check an AgentGarm download's SHA-256 and its minisign signature before you run it. - [Security model](https://agentgarm.com/security): How AgentGarm is built to be trusted: least privilege, hardware-bound approvals, signed everything. - [Privacy](https://agentgarm.com/privacy): What this site and AgentGarm collect, and what they do not. - [Terms](https://agentgarm.com/terms): The licence for AgentGarm's free binaries; terms for this site and the commercial editions are being written. - [Licence and credits](https://agentgarm.com/credits): AgentGarm's licence, and the projects, standards and typefaces it builds on. - [Documentation](https://agentgarm.com/docs): Install AgentGarm, write a first policy, and run it in CI. - [Quick start](https://agentgarm.com/docs/quick-start): Install a policy, run an agent through AgentGarm and see what it did. - [Troubleshooting](https://agentgarm.com/docs/troubleshooting): What you see when something goes wrong with AgentGarm, and the guide step that explains and fixes it, by symptom. - [Install on macOS](https://agentgarm.com/docs/install/macos): Install AgentGarm on macOS from the signed package, then start its services. - [Install on Linux](https://agentgarm.com/docs/install/linux): Install AgentGarm on Debian, Ubuntu, Fedora or RHEL, then start its services. - [Install on Windows](https://agentgarm.com/docs/install/windows): The Windows preview: what it is, and how to install it to evaluate AgentGarm. (preview: built, not yet claimed) - [Install without administrator rights](https://agentgarm.com/docs/install/no-root): Install AgentGarm as yourself from the tarball, and what that gives up. - [Install Central Lite](https://agentgarm.com/docs/install/central-lite): Run Central Lite, the fleet control plane, from its tarball as an unprivileged account. (preview: built, not yet claimed) - [Policies](https://agentgarm.com/docs/policies): What a policy says, and how to move it from audit to enforce. - [AgentGarm in CI](https://agentgarm.com/docs/ci): Govern AI agents in a CI job with the GitHub Action and the dev container feature. - [How-to guides](https://agentgarm.com/docs/how-to): Step-by-step guides for governing AI agents with AgentGarm: policy rollout, approvals, secrets, cost and recovery; CI and Central Lite fleets in preview. ## How-to guides - [Watch your agent for a week, then enforce](https://agentgarm.com/docs/how-to/audit-then-enforce): Start a policy in audit mode, see what enforcing it would have blocked in your own history, fix the rules that get in the way, then switch it to enforce. - [Govern Claude Code](https://agentgarm.com/docs/how-to/claude-code): Install the Claude Code pack, put AgentGarm's hooks in Claude Code's managed settings, run it through the launcher and proxy, and check what is enforced. - [Govern Codex CLI](https://agentgarm.com/docs/how-to/codex): Install the Codex pack, add its hooks to Codex's managed requirements.toml on a root install, start Codex through the launcher, and check what is enforced. - [Govern Cursor, Copilot CLI, Devin, Cline and others](https://agentgarm.com/docs/how-to/other-agents): Install your agent's policy pack in audit mode, put AgentGarm's hooks where that agent reads them, and read what its hooks cannot do. (preview: built, not yet claimed) - [Govern an agent with no hooks](https://agentgarm.com/docs/how-to/wrap): Run an agent AgentGarm has no hook adapter for inside the same sandbox and proxy, and know which of your policy's rules that leaves unenforced. - [Route other model providers through the proxy](https://agentgarm.com/docs/how-to/provider-routes): Send Gemini, Bedrock, Vertex, Azure, OpenRouter, DeepSeek and local-runtime calls through the proxy, accounted, budgeted and checked for secrets. (preview: built, not yet claimed) - [Check that protection is on](https://agentgarm.com/docs/how-to/check-protection): See whether AgentGarm's services are up, what assurance level each agent session has and why, and what to do about a bypass alert. - [Turn on the macOS sentinels](https://agentgarm.com/docs/how-to/macos-sentinels): Activate AgentGarm's Endpoint Security and Network Extension on a Mac, approve them by MDM or by hand, and check that they hold the agent's processes. (preview: built, not yet claimed) - [Get told when an agent is denied or asks](https://agentgarm.com/docs/how-to/notifications): Follow each denial, approval ask and alert as it happens: in a terminal, as desktop notifications, or through your own Linux notifier. (preview: built, not yet claimed) - [Run AgentGarm on Windows](https://agentgarm.com/docs/how-to/windows): Install the Windows preview, start agents in its Job Object sandbox, use the tray, know what Windows enforces next to macOS and Linux, and remove it. (preview: built, not yet claimed) - [See the AI runtimes and desktop AI apps on a machine](https://agentgarm.com/docs/how-to/local-runtimes): List the local model runtimes and desktop AI apps on a machine, see which ones no governed session used, and route local model calls through the proxy. (preview: built, not yet claimed) - [Check for and verify AgentGarm updates](https://agentgarm.com/docs/how-to/updates): Find the newest AgentGarm release, check its signature and checksums before you run it, never go back to an older one, and upgrade in place. - [Allow, ask or deny shell commands](https://agentgarm.com/docs/how-to/command-rules): Write command rules on the parsed command and its wrappers, not its text, choose between asking and denying, and replay commands before you rely on them. - [Limit what an agent can read, write and reach](https://agentgarm.com/docs/how-to/files-and-network): Decide which files an agent may read and write and which hosts it may reach, know which layer holds each rule, and recognise a denial. - [Test a policy before you ship it](https://agentgarm.com/docs/how-to/test-policy): Write what a policy must allow, ask about and deny as a file of events, replay it before you install the policy, and keep both in your repository. - [Approve risky actions with Touch ID](https://agentgarm.com/docs/how-to/approvals): Enrol a Touch ID key on your Mac and approve each command your policy asks about with your fingerprint, for that one call only. - [Approve risky actions on Windows](https://agentgarm.com/docs/how-to/windows-hello): Enrol a security key on a Windows PC and approve each command your policy asks about through Windows' own WebAuthn; Windows Hello itself comes later. (preview: built, not yet claimed) - [Catch poisoned agent configs before launch](https://agentgarm.com/docs/how-to/agent-config-scan): Scan the instruction, settings, hook, plugin and MCP files your agents read for hidden text and auto-approval switches, and gate CI on the result. - [Keep agent versions in check](https://agentgarm.com/docs/how-to/agent-versions): List your installed agents against the versions AgentGarm verified their hooks for, and bring an agent that updated past that range back in range. - [Keep secrets and card numbers out of prompts](https://agentgarm.com/docs/how-to/dlp): See which credentials your agent sends its model provider, have the proxy replace them with cache-safe placeholders, and deny tool calls that carry one. - [Check what agents upload to paste sites and webhooks](https://agentgarm.com/docs/how-to/outbound-inspection): Have AgentGarm read what launched agents send to paste sites, file sharing and webhooks, record or stop secrets, and fix tools that refuse its certificate. (preview: built, not yet claimed) - [Give Claude Code a per-developer gateway key](https://agentgarm.com/docs/how-to/proxy-keys): Give each developer their own gateway key from Central Lite and let Claude Code fetch it through apiKeyHelper; a claude.ai login is never routed. (preview: built, not yet claimed) - [Give agents short-lived AWS, GCP and SSH credentials](https://agentgarm.com/docs/how-to/cloud-credentials): Seal AWS, GCP and SSH credentials in agentgarmd; each task started with agentgarm run gets short-lived ones, and the sealed ones never reach the agent. (preview: built, not yet claimed) - [Run MCP servers through the gateway](https://agentgarm.com/docs/how-to/mcp-gateway): List the MCP servers an agent may use, pinned to a package and hash, and run each through the gateway, where agentgarmd decides every tool call first. (preview: built, not yet claimed) - [Sign in to MCP servers without exposing tokens](https://agentgarm.com/docs/how-to/mcp-sign-in): Give your MCP servers their API keys and OAuth sign-ins from your own terminal; agentgarmd seals each one and the agent never sees a token. (preview: built, not yet claimed) - [Allow only vetted MCP servers](https://agentgarm.com/docs/how-to/mcp-registry): Mirror the MCP registry, have Central Lite sign it for your machines, and let agents launch only servers resolved from that signed snapshot. (preview: built, not yet claimed) - [Contain prompt injection in MCP tool output](https://agentgarm.com/docs/how-to/tool-injection): Strip hidden text from MCP tool results, hold back poisoned tool descriptions, pin tool definitions, and limit what a hijacked agent can do next. (preview: built, not yet claimed) - [Govern browser agents](https://agentgarm.com/docs/how-to/browser-agents): Put AgentGarm's browser shim in front of the browsers your agent launches, so each launch gets a fresh profile and its downloads wait in quarantine. (preview: built, not yet claimed) - [Govern computer-use agents](https://agentgarm.com/docs/how-to/computer-use): Let an agent drive your Mac's mouse and keyboard while AgentGarm's airbag drops its input on password managers, settings and other protected targets. (preview: built, not yet claimed) - [Watch what agents can capture](https://agentgarm.com/docs/how-to/screen-capture): See which agents can capture your Mac's screen, and alert, pause or end a task when a password manager or other sensitive app comes to the front. (preview: built, not yet claimed) - [Set spend budgets and stop runaway loops](https://agentgarm.com/docs/how-to/budgets): Give each agent task a cost budget and a token rate, see what the agent meets at the cap, and stop loops and runaway subagents with limits of their own. - [See what each task cost](https://agentgarm.com/docs/how-to/task-cost): Route an agent's model calls through AgentGarm so each one is counted and priced, and read the spend by provider and model. - [Reconcile spend with provider invoices](https://agentgarm.com/docs/how-to/reconcile-costs): Reconcile a day of this machine's model calls against your provider's own usage and cost reports, read where they differ, and export FOCUS rows. (preview: built, not yet claimed) - [Undo what an agent changed](https://agentgarm.com/docs/how-to/rollback): Preview what rolling back an agent's task would restore and keep, then apply the rollback from the checkpoint taken before the agent started. - [Find out why your agent stopped](https://agentgarm.com/docs/how-to/agent-stopped): Tell whether your agent was suspended, killed, locked out by a kill order or stopped by its budget, get your work back, and find out who can lift it. (preview: built, not yet claimed) - [Trace what one agent task did](https://agentgarm.com/docs/how-to/trace-task): Find an agent's task in the dashboard, read every decision its policy made with the rule behind it, and follow the run from first decision to last. - [Export and verify the audit log](https://agentgarm.com/docs/how-to/audit-log): Export AgentGarm's hash-chained audit log, check it against the node key that signs its heads, and know what a passing or broken check tells you. - [Export evidence for one task](https://agentgarm.com/docs/how-to/incident-evidence): Write one task's signed evidence bundle on its machine, verify it with the node's key, hold its records, and hand it to someone else to check. (preview: built, not yet claimed) - [Send agent events to your SIEM](https://agentgarm.com/docs/how-to/siem): Turn on agentgarmd's exporters: OCSF events to your SIEM, OTel GenAI spans and token metrics to your collector, and notice when an export stops. (preview: built, not yet claimed) - [Send the fleet's security events to Splunk or Sentinel](https://agentgarm.com/docs/how-to/siem-central): Have Central Lite forward the fleet's security and audit events, its own audit log and its alerts as OCSF to Splunk or Microsoft Sentinel. (preview: built, not yet claimed) - [Send a redacted support bundle](https://agentgarm.com/docs/how-to/support-bundle): Write a JSON file of diagnostics with your home directory, user and host names and the secrets AgentGarm recognises redacted, sent only when you share it. - [Govern agents on a headless server or container](https://agentgarm.com/docs/how-to/headless): Check what a server or container offers, set a policy to the headless profile so every ask is denied instead of waiting, and test it before you install it. - [Approve a CI job's risky step from your laptop](https://agentgarm.com/docs/how-to/ci-approvals): Mark the commands a CI job may wait on a person for, and approve the job's request from the console with your passkey, for exactly that call, once. (preview: built, not yet claimed) - [Tour the fleet console in the demo](https://agentgarm.com/docs/how-to/console-demo): Walk Central Lite's console on a fictional company's synthetic data: its machines, an incident, a kill order, costs, outbound inspection and access. (preview: built, not yet claimed) - [Stand up Central Lite](https://agentgarm.com/docs/how-to/central-lite): Set up Central Lite on one Linux host: its org and keys, its HTTPS port, a first admin and a backup passkey, and its first encrypted backup. (preview: built, not yet claimed) - [Keep Central Lite's keys in a hardware security module](https://agentgarm.com/docs/how-to/hsm-keys): Create Central Lite's org with every key wrapped under a key that never leaves your PKCS#11 token, and keep what a restore needs backed up. (preview: built, not yet claimed) - [Sign in to the console with your identity provider](https://agentgarm.com/docs/how-to/console-sso): Let admins open Central Lite's console through Okta, Entra ID or another OIDC provider, with roles from its groups and a passkey still behind every change. (preview: built, not yet claimed) - [Provision and remove people with SCIM](https://agentgarm.com/docs/how-to/provision-users): Let Okta or Entra ID create the people Central Lite knows, give them console roles from their groups, and end their access when they leave. (preview: built, not yet claimed) - [Give people the right access in the console](https://agentgarm.com/docs/how-to/console-roles): Give each Central Lite admin one of six roles, for the whole org or a host group, and know who may request a kill order and who holds each approver key. (preview: built, not yet claimed) - [Enrol developer machines](https://agentgarm.com/docs/how-to/enrol-machines): Enrol a machine in Central Lite with a one-time token that pins Central Lite's key, see it report in the inventory, and let its users enrol presence keys. (preview: built, not yet claimed) - [Deploy AgentGarm with MDM](https://agentgarm.com/docs/how-to/mdm-deploy): Roll AgentGarm out to managed Macs, Linux machines and Windows PCs: the packages, the services, the macOS approvals and the browser extension. (preview: built, not yet claimed) - [Roll out policy from git with canaries](https://agentgarm.com/docs/how-to/policy-rollout): Keep agents.d policies in git and let Central Lite roll each signed commit out to canaries, then the fleet, halting by itself when machines report trouble. (preview: built, not yet claimed) - [Roll out outbound inspection across a fleet](https://agentgarm.com/docs/how-to/outbound-inspection-fleet): List the sites your fleet's agents should not send secrets to, run them in Audit, read what Block would stop, then block through a canary rollout. (preview: built, not yet claimed) - [Stop an agent task on any machine](https://agentgarm.com/docs/how-to/kill-switch): Kill or suspend one agent task from the Central Lite host, check that the node applied it, and revoke a machine you no longer trust. (preview: built, not yet claimed) - [Stop agents across a fleet](https://agentgarm.com/docs/how-to/fleet-kill): Suspend, terminate or lock down agents on one machine, one agent, one person or the whole fleet, with an order two approver keys sign and each node checks. (preview: built, not yet claimed) - [Investigate a task across the fleet](https://agentgarm.com/docs/how-to/fleet-incident): Read a task's timeline in Central Lite with its gaps, export a bundle the org root's evidence key signs, verify it offline and hold its records. (preview: built, not yet claimed) - [Give every agent an owner and an identity](https://agentgarm.com/docs/how-to/agent-identities): Register each person and agent in Central Lite's identity registry, issue and revoke gateway keys, and see which identity every task belongs to. (preview: built, not yet claimed) - [Attribute AI spend by machine and person](https://agentgarm.com/docs/how-to/fleet-costs): Read your enrolled machines' model spend by day, machine, provider and model, and reconcile it against your providers' own usage and cost reports. (preview: built, not yet claimed) - [Set retention, legal holds and data-subject requests](https://agentgarm.com/docs/how-to/retention): Set how long Central Lite keeps each class of record, put a legal hold on the records a case needs, and answer a person's access or erasure request. (preview: built, not yet claimed) - [Prepare a works-council or DPIA pack](https://agentgarm.com/docs/how-to/works-council): Generate templates that describe what your deployment collects, keeps and sends, and switch the fleet to the non-monitoring profile during talks. (preview: built, not yet claimed) - [Back up and restore Central Lite](https://agentgarm.com/docs/how-to/backup-restore): Write Central Lite's encrypted backups by hand and on a schedule, and restore one so that no enrolled node refuses what Central Lite signs next. (preview: built, not yet claimed) The full text of every page: https://agentgarm.com/llms-full.txt