[FEATURE]: persistent daemon agent for workspace context with push/pull coordination #23775

Verification

• I searched existing issues first
• This is a feature/design request, not an implementation PR

Description

I’d like OpenCode to support a persistent background daemon agent that runs alongside the main agent for a workspace.

The main idea is to introduce a first-class background agent that stays alive during the session and can exchange information with the main agent through both pull and push communication.

Core concept

A daemon agent should be able to:

• run continuously in the background for the current workspace
• observe long-lived workspace state
• maintain structured memory/context that should not be recomputed repeatedly
• communicate with the main agent through:
  • pull: the main agent asks the daemon for current state, summaries, or recommendations
  • push: the daemon proactively informs the main agent when meaningful events occur

Why this is useful

The main agent is usually optimized for the immediate task in front of it. During longer sessions, it can lose track of operational context that changes in parallel.

A daemon agent would act like durable workspace intelligence: always on, low-noise, and able to feed the main agent the right information at the right time.

One concrete business scenario: multi-repo workspace tracking

For example, I often work in a top-level project that contains multiple child Git repositories.

In that scenario, the daemon agent could continuously maintain per-repo information such as:

• current branch
• whether the branch has already been merged upstream
• dirty/clean state
• ahead/behind status
• detected language / stack
• how lint / test / build should be run
• environment hints such as “use uv run”, “use make test”, “use pnpm”, etc.

This would help avoid common failures like:

• only noticing very late that the current branch has already been merged upstream and a new branch should have been created
• using the wrong Python command path because the repo expects uv
• repeatedly having to remind the agent how each sub-repo should be linted or tested

Important clarification

The multi-repo Git example is just one concrete use case.

The actual feature request is broader:
I’m asking for a general daemon-agent mechanism with ongoing coordination between background and foreground agents.

Possible UX

Any of these would work:

• automatically start a daemon agent when entering a workspace
• expose a command for the main agent to query daemon state
• allow the daemon to push important events into the main thread
• show daemon-maintained summaries in a side panel or status view

Minimal acceptance criteria

• OpenCode can run a persistent background daemon agent for a workspace
• the daemon can maintain state over time rather than acting like a one-shot helper
• the main agent can pull structured information from the daemon
• the daemon can push important events back to the main agent
• repo/Git awareness can be implemented as one valuable use case of this general mechanism

This would make OpenCode feel much more capable in real polyrepo workflows, where persistent operational awareness matters as much as immediate code editing.