Model freedom
Use suitable local or connected models and change them without abandoning the workspace.
LOCAL-FIRST AI COMMAND CENTER
Your models.
Your tools.
Your direction.
Unicorn Forge brings conversations, projects, agents, skills, computer use, and trusted devices into one observable workspace.
Private pre-alpha · actively building
01 / WHY IT EXISTS
Chat is useful.
Follow-through changes everything.
Forge is being designed for the moment when an answer needs to become action: open the right tool, create the file, run the command, verify the result, and report honestly.
The owner stays in charge through scoped permissions, live activity, approval boundaries, interruption, and durable records of what occurred.
02 / THE WORKSPACE
ONE PLACE TO ORCHESTRATE
Built around the work.
Not a single model, a single cloud, or a single chat. Forge is an environment for choosing and coordinating the right capabilities.
Observable action
Commands, tool results, retries, approvals, and failures appear while agents work.
Personal context
Projects, chats, files, skills, and approved memories stay organized around the individual.
Cooperative compute
Trusted computers can divide suitable tasks and contribute their available resources.
Voice and mobile
Continue conversations and monitor work away from the primary computer.
Extensible tools
MCP servers, skills, computer use, development tools, and services can be managed deliberately.
03 / CONTROL SURFACE
Autonomy deserves instrumentation.
Forge is being built so useful work does not disappear behind a typing indicator.
- 01Plan visibly
Show how a request becomes tasks and which model or machine owns each one.
- 02Act within scope
Use explicit tools and permissions instead of inventing success.
- 03Verify the outcome
Inspect the artifact, command result, or application state before declaring completion.
- 04Recover honestly
Surface stalls and failures with a next action instead of quietly stopping.
04 / PRIVATE FIELD TESTING
Help shape what comes next.
We are preparing a controlled early-testing program for adults who are comfortable with unfinished software, detailed feedback, and clearly disclosed research data practices.