Roadmap

Where herdr-link is going, and what has to be true before it is called 1.0.

What 1.0 means

1.0 is a promise, not a feature count. From 1.0 on, until 2.0:

The Chat protocol v1 is frozen. Adapters written against it keep working; new things arrive only as optional methods and capabilities.
The config keeps loading. config.toml written by any 1.x still works; when a key changes, the old one is still read and doctor says what to change.
The CLI stays put. init, doctor, run, service, hooks: same commands, same flags.
Updating is boring. npm install -g @thl-ai/herdr-link@latest and a restart, never a lost thread or a duplicated room.
It is proven, not just tested: every behaviour in the docs has been used for real, on macOS and Linux, with more than one chat app.

Until then, minor versions (0.x) may still change the protocol and the config, with a note in the release.

0.1 The bridge: spaces → rooms, tabs → threads, replies through agent hooks, permission buttons, new tabs from threads, the archive, commands, the DM control room. Discord adapter. init, doctor, service, hooks.
0.2 Split tabs (several agents in one tab's thread), several machines.
0.3 Docs and this site, doctor --fix, adapters without rooms, no stalls behind Discord's rate limits.
0.4 Telegram (topics as tabs, the bot's chat as the control room); a long turn's progress in one live status message; the first message of a new thread becomes its agent's task; enabled and dashboard switches.
0.5.0 Question forms (Claude Code's AskUserQuestion) answered from the chat; at the desk or away (no pushes while you sit at the computer); a bar in the Herdr window with the link's state and basic settings; herdr-link status, desk, adapter, mod.
0.6.0 Adapters that drive agents, not only chat apps: space.create, thread.created with an agent and a task, room.deleted, thread.status (added before the protocol freezes); install with an agent (/install.md).

What works in tests has to work in a real day.

init installs herdr when it is missing (0.3.4)
Split tabs used live: signatures, replies to a pane, the agent emoji, "Who is this for?"
A rename postponed by Discord's rate limit, seen live and caught up
A clean Linux machine, start to finish: herdr, init, the systemd unit, hooks, a reboot
Restarts on every side: herdr-core, the herdr app, the bridge, the adapter, the computer asleep for a night; nothing lost, nothing doubled
Supported herdr versions written down and checked by doctor
Update from an older herdr-link without losing a thread

The protocol claims to be chat-agnostic. A second adapter proves it, and whatever it finds missing goes in before the freeze.

A Telegram adapter (forum topics as threads, the bot's private chat as the control room) (0.4.0)
init sets up Telegram too
An adapter test kit: a fake bridge that runs an adapter through the whole protocol, so adapter authors know theirs works
Protocol fixes the second adapter asks for, while they are still cheap

More agents with verified keys: Codex CLI, Gemini CLI, opencode
/screen that follows the agent for a while instead of one snapshot
At the desk or away: no pushes while you sit at the computer; a bar in the Herdr window with the link's state and basic settings
The dashboard pinned in the DM
Clearer messages when something is off: herdr app closed, agent signed out, a hook missing

Protocol v1 frozen: schema/chat.openrpc.json marked stable
Config format frozen; anything renamed on the way still read
Security review: owner checks in every adapter path, the hook socket, the adapter environment, what reaches the chat
Docs checked line by line against the code
Used daily for two weeks without touching the code

The 0.9 candidate, with whatever those two weeks turned up fixed.

Ideas, in no order: a read-only room for a client on one project, Slack and Matrix adapters, several owners on one server, a live terminal view.

Something missing here? Open an issue.