What's new in Cerevisor, the local-first app for building and running teams of AI agents. Every update in plain language, with real screenshots from the app.
· v3.8.0
Every agent card now has a lasting chat that is held to the rules of a run; plus connecting the AI tools already on your computer at first launch, connected tools for Claude Code and Codex agents, a run report that opens by itself, and adding a shared workflow to an open world.
· v3.7.1
Factory mode works through a list of development items one at a time, checks each result and commits it locally; plus running workflows from scripts, approvals you can read and change, and new models the day they ship.
· v3.6.0
Keep several worlds open and work in two side by side, start on Home every time, get dialogs that truly block what is behind them, and run agents that get stuck less often.
· v3.5.0
Cerevisor 3.5.0 adds Claude Opus 5.5 and GPT-6 Sol and Luna, lets you give a running Claude Code agent new instructions, lets Codex agents ask you a question, and makes split view keep each pane on its own world.
· v3.4.0
Workflows that move between computers with their folder, one-file sharing with nothing personal inside, decisions you can inspect with inputs that fit, and safer saves.
· v3.3.0
Cerevisor 3.3.0 makes connecting an AI model on a new computer take about two minutes, guides you through your first project, adds quick checks with an optional fast decision model, introduces Free, Pro and Founding member plans, and brings five color schemes.
· v3.2.1
Cerevisor 3.2.1 hardens 3.2.0: Check-by-hand answers can be changed, the Overview says when a run was continued, reviews behave as the guides describe, Continue refuses honestly, and a run's history tells the truth about sizes, secrets and timeouts.
· v3.2.0
Cerevisor 3.2.0 lets you continue an interrupted run where it stopped, have a second agent review any agent's work, see the evidence behind every note, answer an agent's question mid-run, and check a run at a glance from the new Overview on desktop and phone.
· v3.1.1
Cerevisor 3.1.1 puts the agents' conversation on the canvas beside the workflow, lets a new run start from the board its last run ended on, uses your own newer Claude Code install when you have one, and fixes a setting Codex CLI 0.153 no longer accepts.
· v3.1.0
Cerevisor 3.1.0 gives every run a shared board: agents leave notes for each other and for you, ask a named colleague a question and wait for the answer, and you read, post, answer and pin notes live from the run window or your phone.
· v3.0.1
Cerevisor 3.0.1 lets you approve an action a provider's reviewer blocked and run again, labels an agent that ran out of room honestly, keeps re-running one agent under the same checks as a full run, and fixes four bad days: a model that never answers, a shortened tool result, an unreachable server, and money spent by an interrupted Operator action.
· v3.0.0
Cerevisor 3.0 shows you a plain summary before every run, asks again when a workflow gains new powers, keeps a run's status, cost and files open after the app restarts, lets you save a copy of any run, links the Operator to the runs it starts, and stops a run rather than continue unrecorded.
· v2.4.0
Cerevisor 2.4 removes the four-agent ceiling with a scheduler that adapts to your machine and your providers, makes every tool result say plainly whether it succeeded or was refused, keeps agents inside the folder you pointed them at, and adds search to every model picker.
· v2.3.0
Cerevisor 2.3 keeps long runs alive across restarts, adds a searchable World Library, keeps big canvases smooth, reports run results honestly, and recovers from provider outages with fallback, auto-resume, and first-class Grok Build support.
· v2.2.0
Cerevisor 2.2.0 rebuilds MCP support: connect online servers like GitHub and Notion with a normal browser sign-in, add any server by picking, pasting, or typing, and choose exactly which agents get which servers. Plus one-click Antigravity install, Gemini 3.7 Flash as the new default, and an Operator that no longer repeats finished work.
· v2.1.1
Cerevisor now checks a workflow can do what it asks before spending anything, Codex and Claude Code get per-provider approval modes, agent questions become pick-an-option cards, and Operator goals can be edited, parked, completed, and restored, scoped to the world you made them in.
· v2.0.0
Cerevisor 2.0.0 lets Operator run your saved workflows and build its own tools, replaces the provider setup wizard with a single Provider Hub with one-click sign-in for Codex, Claude and Antigravity, adds workflow branching and image support on every agent type, and cuts the app down to exactly two network calls, one of which you can switch off.
· v2.0.1
A same-day patch on 2.0.0: dismissing an Operator card now pauses the objective behind it instead of silently vetoing work you adopted, Operator can see what your existing workflows do so it stops rebuilding them under new names, and a generated workflow can no longer come out with the same agent twice.
· v1.9.0
Cerevisor 1.9.0 adds Progressive Mode, where you describe what you want in chat and it builds and runs it piece by piece, plus Operator, which quietly moves your work forward on a budget while you're away, and a phone app you can now build workflows from.
· v1.8.0
Cerevisor 1.8.0 adds a phone companion for watching and steering runs, standing agent organizations you can chat with, always-on prompt-injection defense, clearer approval modes, and a memory that keeps only what it can prove. Everything runs on your machine, with your keys and data staying local.