Set up Signet

Run setup

signet setup

Setup creates a local workspace when needed, starts or reaches its daemon, and opens a guided modal in the dashboard. Running it again resumes onboarding without resetting your existing configuration. signet configure and signet config open the same flow.

The onboarding flow

1. Welcome explains what memory carries between conversations.
2. Agents offers Claude Code, Codex, Hermes Agent, OpenCode, OpenClaw, Gemini CLI, Pi, Oh My Pi, Kimi, and ForgeCode. OpenClaw requires its Signet plugin package from CLI setup before connecting here. You may connect an agent later.
3. Connection signs in to a provider, stores an API key, or connects a local model. An explicit Test and enable memory action saves the configuration, checks a real model response, and starts automatic processing. Saved credentials can be replaced or signed in again. Failed saves or tests show an error; Set up later continues to sources without enabling processing or claiming the connection works. Existing frozen or shadow controls are respected.
4. Bring your context optionally connects an Obsidian vault or imports files and transcript exports. You can add multiple sources. Indexing and import jobs may continue in the background; their reported state is not a claim that semantic processing has completed. Sources provides progress, errors, retry, and removal.
5. First memory saves a private note under your active agent, then retrieves that same note through scoped search.
6. Ready reports the checks that actually completed. Start a new conversation in your agent to check its integration.

Closing and reopening remembers the UI checkpoint in this browser, but rechecks the model connection. Credentials are stored by the daemon, never in the checkpoint. If browser storage is unavailable, the workspace is still preserved.

Fresh interactive setup leaves identity management off and telemetry disabled. Provider processing waits for the explicit enable action. Existing workspaces keep their settings. Sources and advanced configuration remain available after onboarding.

For unattended systems, use the non-interactive setup reference. JSON plans and flags still use the CLI’s validated setup path; the modal requests existing daemon operations rather than creating a second plan executor. Browser sign-in replaces the retired extractionConnect plan field, which now fails validation.

Identity modes and presets

Identity management controls Signet-owned prompt files. It does not change the memory database or recall features.

• Off: Signet does not create, inject, or sync managed identity files. Your harness-native instructions remain authoritative.
• Minimal: normal startup loads AGENTS.md; DREAMING.md is available only to Dreaming sessions.
• Hermes: uses SOUL.md as primary identity alongside AGENTS.md project context.
• OpenClaw: uses the richer AGENTS, SOUL, IDENTITY, USER, and MEMORY identity stack plus special-session prompts.
• Custom: starts from Minimal and lets you choose startup files explicitly.

Do not copy a fixed workspace tree into a deployment guide. Setup creates agent.yaml, the memory database, and only the identity files selected by the chosen mode and preset. Selected harnesses may create or update their own integration files outside the workspace.

Check and recover

signet status
signet setup

If a browser cannot open, setup prints the dashboard URL with #setup. A remote daemon must already be reachable; local agent connections and Obsidian paths refer to its machine. Use the CLI on your agent’s machine to configure a remote integration.

A partially written agent integration can be retried: the same connector reconciles its files. Failed configuration saves and connection tests are shown in the modal. A saved memory is not saved again merely because recall is still pending. Inspect or remove it in Memory.

Headless setup retains identity presets, network options, Git protection, roster creation, and explicit source flags. Use --non-interactive when supplying configuration changes to an existing workspace; interactive setup will not silently apply those flags. See CLI environment and exit codes for workspace selection.

Source checkouts

Setup and workspace migration do not clone the Signet repository. Saving and recalling memories use the installed application. signet sync and application updates maintain an existing workspace checkout but do not create one.

signet desktop build or signet desktop install explicitly creates the managed checkout when needed for a source build. Contributors can also clone the repository and pass --repo to the desktop command. Existing checkouts, local edits, and branches are preserved; this change does not delete or relocate them. If a managed desktop checkout was removed, rerun signet desktop install to restore its source build path before the next automatic desktop update.