Vocion 5.0

Vocion 5.0 is the biggest change to how the product is organized since workspaces left the repo. Until now, everything a workspace could do sat in one long sidebar. In 5.0 a workspace's work is organized into apps, and every person gets an assistant that works across the workspaces they belong to.

What you can do now

• Install the apps a workspace needs. Every workspace comes with its agents and teams. Turn on Software Factory or GTM from the Marketplace, or in workspace.yaml. Each app shows up on the app rail, with its own workspace picker.
• See everything that needs you in one place. Needs you collects approvals, questions and recommendations from every workspace you are in, your own first. Each row says which agent recommends it, and opens in the workspace it belongs to.
• Start the day from a brief. Your brief combines what is waiting on you with each workspace's latest news.
• Ask once, from your own thread. Your assistant lives in a personal workspace only you can open. Ask it about a deal or a release and it asks the right workspace for you. It names which workspace each fact came from. The cost of that question is charged to the workspace that answered it.
• Work with your assistant by text. Turn on texts and send your assistant a request through Twilio. When it drafts an email, it tells you what will be sent. Reply in plain words. Your reply is read against that one draft, counts only from the number on your profile, and is recorded as your decision. A reply decides only what was raised in that same thread. Slack works the same way once your team sets it up.
• Use Vocion from Claude or ChatGPT. Add a workspace by URL and sign in. The connector uses OAuth in front of the workspace's MCP endpoint.

Earned autonomy, per action

Autonomy is set per action in each workspace, such as updating a CRM record, rather than for an agent as a whole. An action either waits for a person or runs on its own above its confidence bar. The app names six rungs: Observe, Recommend, Assist, Execute with approval, Execute within bounds and Operate autonomously. The first four wait for a person; the top two run on their own above the bar, and every automatic run is recorded.

Small changes that can be undone run on their own above 80% confidence, with Undo, unless your team holds them. Some plugin rules run on their own too, such as merging a pipeline fix.

Every decision a person makes is written to the alignment ledger against the action it covers. When the record shows people keep agreeing with an action, an admin can promote it. A low-risk action needs at least 20 decisions at 90% agreement; a medium-risk action needs 40 at 95%. A high-risk action never moves without an explicit rule in trust.yaml. The server refuses a promotion the evidence does not support. Agreement measures whether people would have let an action run. It does not measure correctness.

If someone rejects or undoes work an action did on its own, it steps down a level and is flagged for review. Sending email and connecting a new data source always wait for a person.

Every run keeps a record

Each chat turn keeps the steps it took, the tools it called with their input and output, the documents it cited, and the model, tokens and cost. Each card keeps who approved it and when. Builds, missions and workflows also record the workspace version that ran. When an agent reads a document, the answer cites it, so you can check a claim against its source.

Deploy on AWS, or on your own host

One OpenTofu apply provisions the network, host, Secrets Manager and DNS in your AWS account, and a bootstrap script starts Vocion with Postgres, Langfuse and TLS. See what to set before you apply. Schedules, workflows and background jobs run inside the app on its own Postgres. Build the Docker image from the repo to run it on any host next to Postgres 16 with pgvector. The whole platform is open source under MPL-2.0.

Make it yours

Write your own agents, teams, skills and rules in YAML, and plugins in code.

Get a demo or read the docs to start.