Skip to content

How to Configure an Agent ​

GSV agents run as processes. Their behavior comes from runtime config, profile context, home context, workspace context, process history, and available syscall tools. Configure the durable inputs rather than editing a hidden prompt.

Set AI Runtime Defaults ​

System defaults live under config/ai/*. Per-user overrides live under users/{uid}/ai/* and win over system defaults for that user.

bash
gsv config get config/ai
gsv config set config/ai/provider openrouter
gsv config set config/ai/model openai/gpt-4.1
gsv config set config/ai/api_key "$OPENROUTER_API_KEY"

Non-root users can set only their own users/{uid}/ai/* keys:

bash
gsv config set users/1000/ai/model gpt-4.1-mini
gsv config set users/1000/ai/max_context_bytes 65536

Sensitive keys such as api_key, token, secret, and password are hidden from non-root system config reads.

Edit System and Profile Context ​

System context applies to every process profile:

text
config/ai/context.d/*.md

Profiles define role-level behavior for process types such as init, task, review, cron, mcp, and app. Profile context is stored as Markdown fragments:

text
config/ai/profile/{profile}/context.d/*.md

Use numeric prefixes to control order:

bash
gsv config set config/ai/context.d/50-local-runtime.md \
  "Use the native gsv target for files in the GSV cloud computer."

gsv config set config/ai/profile/task/context.d/50-style.md \
  "Be direct, inspect files before editing, and explain risky changes first."

System and profile context can use runtime template variables such as identity.username, identity.home, identity.cwd, identity.workspaceId, workspace, devices, mcpServers, and known_paths.

Add Home and Workspace Context ​

Home context applies across a user's processes:

text
~/context.d/*.md

Workspace context applies only when a process is attached to a workspace:

text
/workspaces/{workspaceId}/.gsv/context.d/*.md

Use home context for durable preferences and recurring operating notes. Use workspace context for project-specific instructions, status, and handoff notes. Keep files short and focused; the runtime loads them lexically until config/ai/max_context_bytes is reached.

Configure Tool Approval ​

Tool approval is profile-specific JSON:

text
config/ai/profile/{profile}/tools/approval

Example policy:

bash
gsv config set config/ai/profile/task/tools/approval \
  '{"default":"auto","rules":[{"match":"shell.exec","action":"ask"},{"match":"fs.delete","action":"ask"},{"match":"sys.mcp.call","action":"ask"},{"match":"fs.*","when":{"target":"device"},"action":"ask"}]}'

Rules match exact syscalls or domain wildcards such as fs.*. Conditions can filter by profile, tags, argument prefixes, and target type (gsv or device). Interactive profiles can pause for approval; non-interactive profiles such as cron turn ask decisions into tool errors.

Expose Devices Deliberately ​

Connected devices appear in process context and tool schemas. Agents always see the same tool names (Read, Write, Edit, Delete, Search, Shell); target selects where the syscall runs.

Give devices short notes in the Devices app so agents see why a target exists, not just its id and platform. For example, describe rearden as a Linux home server for GPU work or home automation if that is the routing intent.

Use profile or workspace context to tell agents when a device should be used:

markdown
Use `target: "gsv"` for Kernel files and package state.
Use `target: "macbook"` only for the local checkout under ~/projects/gsv.

Inspect Effective State ​

Useful checks while tuning behavior:

bash
gsv proc list
gsv proc history --limit 20
gsv config get config/ai/profile/task
gsv chat "List your available devices and current working context."

Changes to AI config and context are picked up at the start of the next process run. Reset a process when you want a clean history with the new context:

bash
gsv proc reset