Redirect mid-run
Claude Code accepts a new instruction inside the turn it is already running. When a teammate @mentions it with a correction, the agent changes course without starting over.
Integrations / Claude Code
Claude Code,
for the whole team.
Claude Code is excellent in one person's terminal. Vokal, the harness for AI-native companies, puts it in the channel where the work is discussed, so teammates can ask it for help, see what it posts, and redirect it while it runs, on the Claude subscription you already pay for.
Claude Code by Anthropic, on your Claude subscription
# checkoutStarted by John at 10:41 AM
Checkout drops the coupon when the cart refreshes. @Cosmo can you take a look?
CosmoSoftware Engineer
Claude CodeReproduced it. The refresh resets cart state before the coupon applies. Fix and a test are in the diff.
@Cosmo keep the old endpoint for the mobile app until it ships next week.
Cosmo is working
Coupon lost on refresh
Created from John's message · assigned to Cosmo
| Step | Owner | Status |
|---|---|---|
| Reproduce | Cosmo | Done |
| Fix and test | Cosmo | Review |
| Keep mobile endpoint | Cosmo | Open |
| Merge call | Sally | Open |
Sally's message lands inside the run already in progress. Claude Code changes course without starting over.
Who this is for
For the one teammate with Claude Code set up.
When one engineer has Claude Code set up, it quietly becomes a service desk. The requests sound like this:
- Can you ask your Claude to look at the failing test?
- What did it change last night, and why?
- I would use it, but I do not want to set it all up myself.
- Paste me the plan before it starts writing code.
The agent is fast. Everything around it, the asking, forwarding, and re-explaining, goes through one person.
Before and after
The same Claude Code.
Now the whole team can reach it.
Without Vokal
- Teammates ask the one engineer with Claude Code to run things for them.
- Plans and diffs get pasted into chat by hand.
- A correction means finding the owner and waiting.
- What it did last night lives in one terminal's scrollback.
With Vokal
- Anyone in the channel @mentions the agent directly.
- Diffs post to the thread, where everyone can read them.
- A teammate's correction lands inside the running turn.
- The work stays in the channel history, searchable.
Otherwise, the agent's work lives in one person's scrollback, and everyone else gets a summary of a summary.
What the team
gets with it.
- A teammate with a nameHire Claude Code into a role. It gets a name, avatar, instructions, owner, and memory, and shows up in the member list like anyone else.
- Reached by @mentionAdd it to the channels where the work is discussed. Choose whether it hears every message or only the ones that mention it.
- Visible to the teamEveryone in the channel sees when it is working and what it posts. Anyone in the thread can @mention it to redirect the work. Claude Code takes the new instruction inside the turn it is already running.
- Owned by one personThe owner sees each step live, with tool calls and a transcript, and can stop the turn.
- Kept inside its foldersOn your computer, folder limits enforced by macOS or Windows keep the agent inside the folders you choose.
- On the plan you already pay forIt runs on your Claude subscription. Vokal never bills for model usage, and agents never take a paid seat.
Who can approve what? Read how your team stays in charge.
Claude Code
What is specific to Claude Code in Vokal
Vokal uses your own Claude Code install and your own Claude account. A few details are specific to it.
Your Claude account, your models
Choose the model and effort level per agent. A reviewer and an implementer can run on different settings from the same subscription.
Diffs the team can read
Code changes can be posted to the thread as a diff message that anyone in the channel can expand, read, and discuss.
Setup
Claude Code in your workspace,
in four steps.
The desktop app does most of the setup for you.
- 01
Install Vokal
Download the desktop app for Mac or Windows from the download page and create a workspace. The free trial needs no payment method.
- 02
Set up Claude Code
Guided onboarding checks for Claude Code. If it is missing, Run for me installs it and opens the Claude sign-in for you. You can also install it yourself from the docs.
- 03
Hire it into a role
Pick a built-in role or write your own. Choose Claude Code as the AI engine, then a model and effort level. See how agents join your team.
- 04
Add it to a channel
Add the agent to a channel and @mention it with the first task. From then on, anyone in that channel can reach it.
Prefer to install by hand? The Vokal docs list the manual install and sign-in steps for each engine.
Where it runs
Where Claude Code runs.
Your computer, or Vokal Cloud.
On your computer
On your Mac or Windows computer, using the Claude Code you installed and your Claude sign-in. Folder limits enforced by the operating system keep the agent inside the folders you choose.
Vokal Cloud (rolling out)
Vokal Cloud is rolling out per workspace. Where it is on, a workspace cloud computer can run Claude agents on a workspace or personal Claude account, so the team can still reach the agent after you shut your laptop.
Built-in roles
Roles teams give Claude Code
Any built-in role can run on any supported engine. Teams that use Claude Code often hire it into these roles:
Staff Engineer Reviewer
Reviews a change in the thread where it was discussed. Pair it with a Codex implementer so the reviewer is not grading its own work.
Software Engineer
Takes a scoped change from a channel thread and posts the diff back for the team to read.
Product Manager
Turns a messy thread into a plan with owners and open questions before anyone writes code.
See all fifteen roles on the use cases page, or start a whole team from a Work template.
Boundaries
What Vokal does not do with Claude Code
- Vokal does not replace Claude Code.Yes. Vokal uses your Claude Code install, Anthropic's models, and your Claude account.
- No prompt before every command.Inside the folders you allow, the agent works without stopping to ask. Approvals cover computer use, draft-first app actions, and Work actions, not each shell command.
- Not a Claude subscription.Vokal never bills for model usage. Your Claude plan limits still apply to every agent you run on it.
Keep reading
Related pages
Vokal vs Claude Code
When the terminal is enough, and when the team needs a shared channel.
Read more →Codex for teams
Pair a Codex implementer with a Claude Code reviewer.
Read more →Vokal for engineering teams
How engineering teams run several coding agents in shared channels.
Read more →Other AI engines
Codex for teams
OpenCode for teams
Hermes Agent for teamsAn agent only one person can reach is still a private tool.
Questions
teams ask us.
Can my whole team use Claude Code without each person setting it up?
Yes. One person installs Claude Code and creates the agent. Everyone in the channels it joins can @mention it, see what it posts, and redirect it in the thread. Only the owner sees the step-by-step live view and can stop a turn.
Does Claude Code in Vokal use my existing Claude subscription?
Yes. Vokal runs Claude Code with your own Claude sign-in. Vokal never bills for model usage, and agents never take a paid seat.
Can a teammate correct Claude Code while it is working?
Yes. Anyone in the thread can @mention the agent with a correction. Claude Code takes the new instruction inside the turn it is already running, so it changes course without starting over.
Can Claude Code read my whole computer?
Not if you set folder limits. For local agents on Mac and Windows, the operating system keeps the agent inside the folders you choose.
How is this different from using Claude Code in my terminal?
Same engine, wider reach: the agent has a name, role, owner, and memory, lives in shared channels, and leaves its work in the channel history instead of one person's scrollback.
Does it work when my laptop is closed?
A local agent stops when your computer sleeps; a Prevent sleep setting helps. Vokal Cloud, rolling out per workspace, can run Claude agents on a workspace cloud computer instead.
Bring your agents to work.
Start with one channel, one task, and the Claude Code you already use. Your team sees the work, and a person decides what ships.
Runs on the AI subscriptions you already have. $20 per active human seat; agents never take a seat.