Examples

The setups you'll reach for most. Each guide linked here goes deeper in the fhold docs.

A morning briefing

fhold task create project-news --schedule '0 8 * * 1-5' --prompt 'Check project news and save a report'
fhold task show project-news
fhold task run project-news        # try it now
fhold task history project-news

Reports land in knowledge/inbox/, where the agent can read them back to you. pause and resume toggle a task; remove keeps the definition in knowledge/disabled-tasks/. See Managing fhold.

Connect Claude Desktop

Claude Desktop reaches a local fhold through a desktop extension, since its remote-connector form needs a public https:// address. The extension is a plain MCP bridge; Guardian still does the authenticating:

fhold guardian enable
fhold credential add claude-desktop read
fhold connect claude --credential claude-desktop

Install the .mcpb extension under Settings → Extensions → Advanced settings, then paste the printed address and key. Start with read and move to full only if Claude should change the workspace. See Claude Desktop.

Put it on Discord

fhold portal token discord --bot-token-file -
fhold portal access discord --users 123456789012345678 --no-apply
fhold credential set-policy discord chat
fhold portal credential discord --credential discord --no-apply
fhold portal enable discord
fhold status

Every non-empty allowlist must match, and an empty allowlist refuses everyone, so name yourself explicitly for a personal bot. chat is the safe default for a shared platform. Slack is the same flow with slack, a bot token plus an app token, and --channels. See Discord and Slack.

A second instance

fhold install --name work-agent          # lands in ~/fhold/instances/work-agent
FH_HOME=~/fhold/instances/work-agent fhold setup

Each instance under ~/fhold/instances/ is its own Compose project with its own knowledge, credentials, workspace and ports; fresh setup picks free ports automatically. Commands act on ~/fhold/instances/default unless FH_HOME names another instance's folder, so keep it explicit for every command meant for the second one. In Admin, Create new instance does the same.

Back up and move it

fhold backup --to /private/path/backup

fhold install --name restored --no-start
FH_HOME=~/fhold/instances/restored fhold restore --from /private/path/backup --dry-run
FH_HOME=~/fhold/instances/restored fhold restore --from /private/path/backup --apply

The dry run lists what will be restored and what needs a separate opt-in (provider auth, private environment, portal maps). Backups are unencrypted, so keep them private. See portable backup.

Let Claude Code or Codex work in the workspace

Experimental in the alpha: the Assistant image carries native Claude Code and Codex workers that share the agent's /work directory. They are separate coding agents, not clients of the fhold agent, and they keep their vendor's own sign-in, trust and approvals:

fhold remote enable codex --sandbox read-only
fhold remote status codex
fhold remote pair codex

Nothing is pre-trusted and no public URL, port or SSH daemon is added. See native remote workers.