Workspace and identity

Workspace selection

A Signet workspace owns its configuration, identity files, database, daemon state, and generated artifacts.

Resolution is, in order:

An explicit CLI --path value.
SIGNET_PATH for the current process.
The persisted workspace selection.
~/.agents/.

Inspect or change the persisted selection with:

signet workspace status
signet workspace set /path/to/workspace

Use SIGNET_PATH for a one-off daemon, test, or service invocation. Do not point two writable daemons at the same workspace.

Workspace files

File or directory	Ownership
`agent.yaml`	Operator configuration.
`AGENTS.md`, `SOUL.md`, `IDENTITY.md`, `USER.md`	Agent and user context. Preserve these as source material.
`MEMORY.md`	Generated working-memory projection. Do not hand-edit it.
`DREAMING.md`, `HEARTBEAT.md`, `BOOTSTRAP.md`	Prompts for their named special flows, not ordinary startup context.
`memory/memories.db`	Daemon-owned SQLite state. Do not edit with SQL while the daemon is running.
`.daemon/`	Runtime state, logs, auth material, and telemetry audit data.
`.secrets/`	Encrypted secret storage. Keep it out of source control.

The configuration loader accepts the first present of agent.yaml, AGENT.yaml, or config.yaml, in that order. Prefer agent.yaml for new workspaces.

Minimal baseline

signet setup produces the right starting point for the installed release. A small hand-maintained configuration can look like this:

agent:
  name: "My Agent"
  description: "Personal assistant"

embedding:
  provider: ollama
  model: nomic-embed-text
  base_url: http://127.0.0.1:11434

search:
  alpha: 0.7
  top_k: 20
  min_score: 0.3

memory:
  pipelineV2:
    telemetryEnabled: true
    autonomous:
      enabled: true
      frozen: false
      maintenanceMode: execute

auth:
  mode: local

network:
  mode: localhost

This is a baseline, not a schema dump. Add only the options you operate. Configure model targets and workload bindings in Inference and routing, not by reviving legacy memory.synthesis fields.

Embeddings and search

The embedding provider, model, dimensions, endpoint, and optional api_key belong under embedding. Supported embedding providers include native, llama.cpp, Ollama, and OpenAI-compatible remote providers. Store remote credentials through Secrets and reference them with $secret:NAME rather than writing a key into YAML.

Changing an embedding profile initiates an index migration. Confirm its progress from the daemon status and embedding status endpoints before retiring the old provider or model.

search.alpha controls hybrid retrieval weighting. top_k and min_score bound candidate collection and returned results. Change search tuning deliberately and verify representative recall queries after restart.

Applying a change

Most long-running workers receive configuration at pipeline start. Restart after changing embedding, inference, pipeline, auth, or network settings:

signet daemon restart
signet daemon status --json
curl -fsS http://127.0.0.1:3850/health/ready

If a configuration error prevents startup, use the exact daemon error as the migration guide. Do not delete legacy keys blindly; preserve a backup of agent.yaml, make the smallest correction, then restart.