Data and Network

Monad stores its state on your own machine and sends no telemetry, analytics, crash reports, or usage pings. Requests to the model providers you configure do leave the machine, because that is where inference happens. What Monad stores, and what leaves your machine. Every claim here is a statement about the current code — if you find a discrepancy, that is a bug worth reporting.

The short version

Monad sends no telemetry, analytics, crash reports, or usage pings. Ever. There is no opt-out because there is nothing to opt out of.
No account is required. Monad has no sign-up, no license check, and no server of its own that it phones home to.
All state — sessions, transcripts, memory, credentials, settings — lives on your disk under ~/.monad.
The daemon binds loopback only by default. A bound loopback port is not an exposed port.
The traffic that does leave your machine is traffic you asked for: your model provider, and whatever tools you enable.

What leaves the machine, and when

Nothing in this table happens until you configure or invoke it.

Destination	Triggered by	Carries
Your model provider’s API	Every agent turn	The prompt: system prompt, transcript, injected memory and skill text, tool results
`models.dev`	Model settings, pricing and tier resolution	Nothing about you — it fetches a public catalog
`api.github.com`, `registry.npmjs.org`	`monad atom install`, `monad skill install`, MCP binary install	The package coordinates you asked for
Glama / Smithery / the MCP registry	`monad mcp search`	Your search query
GitHub Releases	`monad update` (and only then)	Nothing but the release request
Qdrant’s GitHub release	First use of the optional `mem0` memory backend	Nothing — a binary download
MCP servers you configure	Their tool calls	Whatever the tool call contains
Channels you configure	Inbound and outbound messages	The conversation on that platform
Peers / Monadix, if configured	Delegated subtasks	The instruction you delegate

There is no background update check. monad update contacts GitHub only when you run it. The model provider is the significant one: an agent turn sends the model whatever is in its context window. If a file’s contents were read by a tool, they are in the transcript, and the transcript goes to the provider on the next turn. Choose your provider accordingly — that is a property of using a hosted model, not of Monad.

Observability is off by default

Monad is instrumented with OpenTelemetry, but the exporter has no default endpoint:

// config.json
"observability": { "endpoint": "" }   // empty = disabled, the default

Set it to an OTLP endpoint and traces and metrics are exported there — to a collector you run. Nothing is sent to Monad or to any third party. In development mode the endpoint auto-defaults to http://localhost:6006 (a local Phoenix instance) unless you set it yourself; that is still your own machine.

Where your data lives

~/.monad/
  configs/config.json         settings
  configs/agents.json         agents, providers, and their credentials
  configs/mesh.json           mesh agents, ACP agents, peers, Monadix
  credentials/auth.json       agent runtime credentials (mode 0600)
  db/monad.sqlite             sessions, messages, events
  memory/                     durable facts, knowledge graph, laws
  atoms/                      installed packs, skills, MCP configs, locales, providers
  agents/<agentId>/           per-agent workspace
  workplace/<projectId>/      managed Workplace Project data
    shared/                   all project sessions and members
    agents/<projectMemberId>/ one member across project sessions
    sessions/<sessionId>/     all members in one session
    runtime/<sessionId>/<projectMemberId>/
                              one member-session runtime
  logs/                       daemon logs

On Linux these paths split across the XDG base directories. MONAD_HOME overrides the complete data root. For managed agents, Monad grants the provider the four concrete shared, member, session, and runtime directories that belong to that runtime. Monad does not grant the containing project directory as one broad working root. Provider sandbox enforcement still depends on the selected provider and launch mode; these boundaries do not replace host operating-system permissions. Files in a session workspace are collaborative and writable by every managed agent in that session. Product rules can be narrower. For example, Kanban requires its Product Design and Tech Design documents to be published by the assigned host from the task’s canonical session directory; that check is enforced by the Experience API rather than by a filesystem ACL.

Secrets

Credentials are never returned by the API. Settings responses mask them; monad credential list shows only a preview.
Secrets are never written to logs, transcripts, or structured CLI output, and never passed in argv.
auth.json is written mode 0600 and sits inside a vault that the agent’s own filesystem tools are denied — an agent cannot read its way to your keys.
Agent Runtime Credentials go further: generated code receives a per-run sentinel, and the real value is substituted only on the leg to the hosts you allowed. See agent-runtime-credentials.md.
Windows has no chmod; the daemon relies on per-user profile ACLs there. Documented rather than pretended-away.

Third-party code

monad license list lists every third-party package in the production dependency graph with its license. Atom packs, skills, and MCP servers you install are third-party code that runs with the access you grant it — treat them like installed software, and read SECURITY.md before installing something you did not write.

Documentation site data

The claims above describe the Monad product runtime. A hosted Mintlify documentation site is a separate web service. It may process page views, searches, page feedback, or AI assistant requests only when the project owner enables the corresponding Mintlify Dashboard add-on. The documentation site does not automatically receive sessions, credentials, or configuration from your local daemon. Do not paste secrets, private transcripts, or sensitive logs into feedback.

Deleting your data

monad purge sessions     # clear transcripts, keep configuration
monad purge all          # clear sessions, config, auth, and usage
monad purge              # wipe and rebuild ~/.monad entirely (double confirmation)

To remove Monad itself as well, see Uninstalling.

Get Started

Monad Mesh: The Agent Team Runtime

Monad Agent Runtime: The First-Party Agent

Operate the Daemon

Trust and Policy

Reference

Data and Network

The short version

What leaves the machine, and when

Observability is off by default

Where your data lives

Secrets

Third-party code

Documentation site data

Deleting your data

​The short version

​What leaves the machine, and when

​Observability is off by default

​Where your data lives

​Secrets

​Third-party code

​Documentation site data

​Deleting your data

The short version

What leaves the machine, and when

Observability is off by default

Where your data lives

Secrets

Third-party code

Documentation site data

Deleting your data