optio-opencode 0.4.0

Run opencode web as an optio task; local subprocess or remote via SSH.

optio-opencode

Run opencode web as an optio task — local subprocess or remote over SSH — with opencode's UI reachable through optio's UI components.

What it does

Given an OpencodeTaskConfig (workdir contents, prompt, deliverable callback), optio-opencode:

1. Provisions a fresh workdir on the chosen host (local or remote).
2. Writes AGENTS.md (base prompt + your instructions) and opencode.json (your config) into it.
3. Installs the opencode binary if missing (remote mode only).
4. Launches opencode web with a random auth password.
5. Registers the opencode UI as a widget that optio's UI components can embed via the widget proxy — SSH tunnel hidden from optio-api.
6. Tails a log file the LLM writes to and translates structured lines into optio events:
   • STATUS: … → ctx.report_progress(percent, message)
   • DELIVERABLE: <path> → fetches the file, invokes your on_deliverable callback
   • DONE [summary] → clean completion
   • ERROR [message] → failure
7. Cleans up workdir and SSH connection on teardown.

The same OpencodeTaskConfig works for local and remote modes; only SSHConfig differs.

When to use it

You want an opencode-driven assistant session as a managed optio task — surfaced through optio's UI, with progress reporting and file deliverables — without writing the host management, log parsing, or widget plumbing yourself.

Installation

pip install optio-opencode

Python 3.11+. Depends on optio-core, optio-host, and asyncssh.

Minimal example

from optio_opencode import create_opencode_task, OpencodeTaskConfig
from optio_host import SSHConfig

config = OpencodeTaskConfig(
    workdir_files={"AGENTS.md": "Do the thing.", "opencode.json": "{...}"},
    on_deliverable=lambda ctx, path, text: print(f"got {path}: {len(text)} bytes"),
    ssh=SSHConfig(host="worker-1", user="optio", key_path="~/.ssh/id_optio"),
)

task = create_opencode_task(config)
# Schedule / run via optio-core as usual.

Set ssh=None for local subprocess mode.

Messages

• use_client_messages (bool, default False) — enable the CLIENT_MESSAGE: log keyword: the agent can push {keyword, data} messages to the browser session that launched the task (surfaced via optio-ui's onClientMessage).
• on_caller_message (async callback, default None) — enable the CALLER_MESSAGE: log keyword: the agent can push {keyword, data} messages to your application. Signature (hook_ctx, keyword, data) -> str | None; a non-None return is sent back to the agent as feedback. Keywords that are not enabled are absent from both the parser and the agent-facing protocol documentation.

License

Apache-2.0.