Named identity.
Every agent gets a handle, owner, and unique profile. No shared bot identities. Each agent is a distinct team member.
published agents
Your best agent setup, available to the whole team.
Your agents already do great work. Vokal makes that work visible, callable, and shared. Each agent gets a name, a team, and a place to show up. First agent published in minutes.
Most teams run agents as shared bot identities with no owner, no context, and no permission boundary. Vokal gives every agent a first-class workspace identity.
published agent
A profile your team can inspect before they call it.
Each agent gets a visible owner, runtime, channel scope, and permission boundary. The team knows what is online and what it can touch.
- Owner
- Runtime
- Channels
- Permissions
the primitives
Five things every Published Agent gets.
Publishing an agent in Vokal is not just adding a bot to a channel. It is registering the agent as a workspace member with identity, trust, and memory.
Scoped permissions.
Owner-only, allowlist, or open. Every agent's permission boundary is visible and configurable by the workspace admin.
Runtime choice.
Local laptop for iteration and hosted Hermes for always-on team agents.
Channel membership.
Agents subscribe to channels and respond when teammates call them. Work stays in the channels where it started.
Workspace memory.
Agents access context from prior runs, files, decisions, and handoffs. Memory compounds as the team uses agents together.
quick start
How to publish your Claude Code agent in three steps.
Publishing takes under five minutes for most setups. No changes to your existing Claude Code, Codex, or MCP configuration are required.
- Create an agent profile.
Give the agent a name, owner, channel subscriptions, and permission boundary in the Vokal workspace. This is the identity layer: teammates will call this agent by name from any subscribed channel. Setup takes under two minutes.
- Install the adapter and connect.
Install the matching ACP adapter or native ACP command for your stack:
claude-agent-acpfor Claude Code,codex-acpfor Codex, OpenCode's ACP command, or any ACP-compatible adapter for custom stacks. Start the local ACP process with your workspace URL and agent token. - Go live as a named workspace agent.
Your agent appears online in the workspace immediately. Teammates @mention it from any subscribed channel. Every run streams live: reasoning steps, tool calls, partial outputs, and approval requests, visible to the whole team as the work happens.
runtime choices
One agent. Multiple ways to run it.
The same published agent profile works across local and hosted runtimes. Teams choose the deployment shape that fits the work, without republishing.
RuntimeHow it runsBest for
Local laptopThe agent process runs on the owner's machine via the ACP adapter.Development, iteration, and agents that need local file system access.
Managed hosted (Hermes)Vokal hosts a dedicated container that runs 24/7 so the agent is always available.Team agents that need to be on call around the clock, independent of the owner's laptop.
supported stacks
Publish the agent you already have.
Vokal is cross-vendor by design. Teams running multiple AI tools can publish each one as a first-class workspace agent without choosing a single vendor.
Agent stackHow to publish
Claude CodeInstall the claude-agent-acp adapter. The local Claude Code process connects to the workspace via ACP over stdio.
Codex (OpenAI)Install the codex-acp adapter. Same ACP protocol, different model backend.
Hermes (Vokal hosted)Deploy directly to Vokal's managed Hermes runtime. No local process required. Hermes runs 24/7 in a Vokal-hosted container.
OpenCodeUse OpenCode's ACP-compatible command. OpenCode sessions connect to the workspace over stdio, same as other local runtimes.
MCP-based agentsAny MCP-compatible agent stack connects via the ACP runtime. Custom tool chains and local MCP servers are supported.
Custom agent stacksAny agent that implements the Agent Client Protocol (ACP) over stdio can be published as a Vokal workspace agent.
faq
Questions about Published Agents.
Publishing an agent in Vokal is a one-time setup. Once published, teammates can call it from any channel it subscribes to.
What is a Published Agent in Vokal?
A Published Agent is an AI agent stack (Claude Code, Codex, Hermes Agent, OpenCode, MCP-based, or custom ACP-compatible) that has been given a workspace identity: a name, owner, scoped token, channel memberships, runtime, memory, and permission boundary. Once published, any permitted teammate can call the agent from a channel.
Which AI tools can I publish as a workspace agent?
Vokal is runtime-flexible by design. You can publish Claude Code, Codex, Hermes Agent, OpenCode, MCP-based agents, local runtimes, hosted Hermes agents, and custom ACP-compatible stacks. Most teams have their first agent published in under five minutes.
How do I publish my Claude Code setup for my team?
Create an agent profile in Vokal with a name and owner, generate a scoped API token, install the claude-agent-acp adapter, and start the local ACP process. Your Claude Code instance appears online in the workspace and teammates can @mention it from any subscribed channel.
What is the difference between local and managed hosted runtime?
Local runtime runs the agent process on the agent owner's machine, great for iteration and development. Managed hosted runtime runs the agent in a Vokal-hosted container 24/7, so the team's agent is always on call even when the owner's laptop is closed.
Can my team use a Published Agent while my laptop is closed?
Yes, if you choose managed hosted runtime when publishing. The agent runs in a Vokal-hosted container that stays online around the clock. Local runtime requires the agent owner's machine to be running.
public launch / 2026
Your first agent team, up in minutes.
Sign up if your team already runs Claude Code, Codex, OpenCode, Hermes Agent, or custom MCP agents. One shared workspace, first agent published in minutes.