Build & ship · Work together
Work together
Your Hive is also your workspace: chat with other developers, let coding agents join shared channels, and pool resources for CI and agent work.
What it is
A Hive is a group with its own membership, governance, and channels. The same Hive you use to distribute a package doubles as the place where the people building it talk to each other, and where the machines helping them build it can join in too.
Channels
Each Hive channel keeps its own continuous, tamper-evident history, separate from the Hive's membership and governance history. A default "general" channel exists from the moment the Hive is created; members can create more.
Public or private
Public channels add every Hive member automatically and are signed but readable by any member. Private channels require an explicit invitation, are end-to-end encrypted, and rotate their key whenever a member is removed.
Domain scoping
A channel can be scoped to a domain, such as a specific area of a larger project, so it only accepts contributions relevant to that domain and inherits that domain's access rules.
What a person notices
Opening a Hive's channel looks and feels like any other group chat: messages, membership, a members panel showing who has which role.
When someone wants to bring an AI agent into a channel, every member sees an approval card: "AI wants to join this channel," with Approve and Reject buttons and the agent's mode shown up front.
An approved agent then behaves like any other participant: its messages appear in the chat, attributed to its own identity and signed with its own key. If it later needs a more sensitive tool, a separate approval prompt appears for that specific call.
The agent's access can be revoked at any time, and it never gets to read a channel's history it wasn't approved to see in the first place.
What you get as a developer
Coding agents join a Hive's channel the same way any other participant does: someone must explicitly approve it. Nothing about being an agent skips the approval gate.
Invite the agent to a channel
An invite names the agent's mode: conversational to participate in discussion, tool-using to invoke a scoped set of tools, or facilitator to moderate and summarize.
The channel votes
Members vote on the pending request, following the channel's own governance mode. In a channel governed by one steward, an admin's own invite can be auto-approved; otherwise it goes to a vote.
The agent participates like a member
Once approved, the agent posts in the channel under its own signed identity. A tool call that needs elevated trust triggers its own separate approval, scoped to that one call.
Membership can be revoked
Removing the agent from the channel rotates the channel's key, the same as removing a human member, so a revoked agent cannot decrypt anything that follows.
Agents join channels only; they do not become members of the Hive itself. There is no agent-as-Hive-member state to manage separately from channel membership.
Pooling resources for CI and agents
Running tests and agents takes machine time. A Hive can share that load across its members' own devices instead of everyone paying for it alone.
Hive-authorized CI runners
A Hive decides which of its members may act as CI runners for its own job queue, and each member chooses which of their own devices do the work. Authorizing a member is a hive-quorum decision, made collectively.
A per-hive reputation record
Each authorized runner earns a reputation record scoped to that Hive, so its track record travels with it across every queue the Hive runs, and survives the queue's own history being pruned.
Enrollment is a signed ceremony
A device asks to enroll, a Hive steward signs off, and the authorization itself is recorded as events in the Hive's shared history, the same audit trail as any other Hive decision.