Configuration

An administrator changes server settings in the dashboard. Each member sets their own machine’s capture folders there too. Changes apply through the server; you do not need to hand-edit a configuration file on every laptop.

Settings

Settings has five sections for the whole server.

Myco’s work

• When Myco works controls Work on a schedule, how long projects count as active, and when a quiet session is ready for a title. Scheduled work starts off. Title imported sessions separately enables titling older sessions.
• How much at once controls Tasks at once, Runs of one task at once and Runs of one task per hour. The default for runs of one task at once is 2 across all projects. Further runs queue. Clear an override to restore the default, or choose No limit.
• Which agent does the work sets Preferred agent and Then try, in order, among agents the registered runners offer.
• What sessions receive controls Session-start instructions, injection at session or subagent start, spores and plans with prompts, and the planning reminder. The instructions are one shared template for all projects, up to 4 KB. Items per prompt shares its count between spores and plans; a 300-token budget can serve fewer.
• Code map controls Update the map on its own, Update every and Paths left out. Automatic updates start off; when enabled, their default interval is six hours. They also need scheduled work, the project’s Code map switch and a connected repository.

First extraction of a large history ramps at two automatic runs a day until the initial history drains or the active window expires. The normal automatic ceiling is 12 learning runs per project per day, with three slots reserved for recent live sessions. How Myco learns explains the four kinds of work.

Models and keys

Claude Code tiers, Codex tiers and OpenCode tiers choose models and effort for Myco’s tasks. Configure the tiers a runner will use; unset Codex or OpenCode models hold that work until configured. Per-task overrides can choose a different agent, tier, model, effort or schedule for one task. A model pin also needs an agent in the same override.

Runners use agents already signed in on their machines. Registering a runner is separate from member sign-in; choosing a preferred agent does not register one.

Search embeddings sets the provider and model used to search by meaning. Cloudflare servers default to Workers AI. A local server has no default embedding provider and matches words until one is configured. Changing the embedding model rebuilds the search index in the background while search remains available.

Capture and retention

New repositories › Create a project for it controls whether a repository inside a machine’s capture folders can create a project when none holds its remote. If off, it waits for an administrator to connect it. Repositories whose remote already belongs to a project can still join that project.

Importing past sessions controls Import past sessions on join, how far back to reach and how many sessions to bring per agent. Turning import off also refuses manual import.

What Myco keeps controls raw content and task record retention. Keep raw content hot for defaults to 90 days; older raw bytes move into verified archive storage. Sessions, prompts, replies, tool calls, plans and spores stay indefinitely.

Backups

Backups controls the interval and retained recovery copies. A local server writes verified recovery copies; a Cloudflare server stages a copy for an operator to complete. Store checks controls automatic optimization and integrity checks. Their results appear on Health. For operator commands, see Run your own Myco.

Sign-in and access

This section points to People & machines for membership and machine access, and Project settings › Access keys for external agents. Keys are scoped to one project; see External agents.

What Myco does in each project

An administrator changes these in Project settings › What Myco does here. All three switches are on by default; an explicit opt-out stays off.

Map work and learning from code need Project settings › Repository connected. Agent work needs a registered runner. Scheduled work needs its separate server switches. Archived projects are excluded from automatic work.

Starting work by hand

Every member can use Myco’s work › Run a task. By default a member who is not an administrator can start each task four times in a rolling day, across all projects. Administrators have no member daily limit.

To change that allowance for one task, set memberRunsPerDay in Models and keys › Per-task overrides › Task overrides. For example:

{"extract-curate": {"schedule": {"memberRunsPerDay": 10}}}

This gives each member ten manual learning starts a day. It is separate from the automatic schedule’s daily ceiling. Concurrency limits queue work; reaching a member daily ceiling refuses another start until the allowance returns.

What you set per machine

Open My machines from the account menu and choose Its settings:

• Folders it captures names parent folders whose Git repositories join automatically; the default is ~/Repos.
• Extra plan folders adds plan locations beside each agent’s own plan folder. Entries may be relative to a project, start with ~/, or be absolute.

Only the member who owns a machine can change its settings. A change is picked up at that machine’s next session start. For a repository outside the capture folders, run myco member join from that repository.

To inspect effective settings from a connected repository:

myco config get

It shows server, machine and member settings. The configuration CLI is read-only in member mode; use the dashboard to change settings. Personal settings such as a log level may be shown without being consumed by the 2.0 runtime.

Sandboxes and CI

Generated hooks do not sign a sandbox or CI job in.

One piece already works: generate an agent’s settings with Myco’s hooks, without signing in, for a job to write into its sandbox:

myco settings --harness claude-code --project <project-id>