Frequently asked questions
The short version of how everything fits together — the guides below go deeper on any part of it.
Every part of Frontier, explained
Grouped by topic, each starting with the basics before working up to the more advanced pieces.
Getting Started
What is Frontier?
The short version: one place for your ideas, repos, and agents to live together.
Picking your context
First launch asks who you are, not for a password — here's why, and what the setup walks you through.
Your first week with Frontier
A rough on-ramp: what to try on day one, and what to reach for once the basics feel automatic.
Sharing your setup with everyone
Put a frontier.json in a public GitHub repo, and anyone can review your agent team and install it in one click.
Settings that apply everywhere
The handful of baseline choices every workspace inherits — where work lands, and which coding CLI runs it.
Customizing the prompts behind every run
The instructions Frontier wraps around your messages are editable — and worth reading once before you change them.
Workspaces & Threads
Workspaces & Threads
Your project's home base — repos, conversations, and history, all attached to one place.
Diffs, history, and editing files directly
Every change a conversation makes stays visible — as an inline diff, a running history, or a file you can open yourself.
Organizing bigger projects with phases
A big initiative doesn't have to live in one workspace — nest smaller ones underneath it instead.
One-off conversations
Sometimes you just want to ask something — a floating panel skips creating a workspace entirely.
Revision history
Every edit to a workspace, domain, or agent is recorded automatically — what changed, from what, to what.
Replies you can use, not just read
An agent can answer with a table, cards, a form or a live view, with buttons that act on what it found.
Charts in a thread
Ask for a picture of the data, and the agent draws it right under its reply.
Domains
Domains
Write down what your codebase needs an agent to know, once, and stop repeating yourself.
Keeping Domains and repositories in sync
A Domain's notes only help if its workspace can actually see the code they describe.
Writing a Domain that actually gets used
The difference between a Domain that helps and one nobody reads comes down to a few habits.
Agents & Conductor
Agents & the Conductor
Two different kinds of automation — a one-off nudge, and something that keeps going on its own.
Tuning how often an agent checks in
How often an agent looks, and how long it's allowed to run, are both yours to adjust.
Reading the Audit Trail
When an agent didn't do what you expected, this is the first place to look — not the last.
Agents that run code, not a prompt
Some of what a loop does is mechanical. A Script agent does that part directly, with no model run at all.
Letting code decide when an agent runs
A few lines in front of the prompt, so the model only wakes up for the threads that are actually worth it.
What a script is allowed to touch
Nothing, until you say otherwise — and everything it does say is written down.
Sharing one resource between agents
When only one agent can hold the wheel, they take turns through a lock that expires on its own.
Remembering things between runs
A small set of saved values per workspace, so a loop knows what it already did.
Strategies
Reusable Strategies
Save a multi-agent pattern that worked, and bring it to life again with one click.
Root roles vs. their own workspace
The one real judgment call when designing a role: does it need a space of its own?
Referencing sibling roles in a prompt
One role's instructions can point at another by name — resolved automatically the moment you instantiate.
Forking the built-in Watcher/Addresser and Engineer/Reviewer strategies
Two proven patterns ship ready to copy — start there before designing one from scratch.
Memories
Connectors
Connectors
Frontier speaks Claude Code, Codex, Gemini, and Grok natively — and leaves the door open for other coding agents too.
Choosing a connector per workspace
Most people never touch this — it's there for the rare project that wants something different.
Running a model on your own machine
Install an open model from Hugging Face and it runs here, on your hardware, for free.
What "scaffolded" means for a connector
Aider is listed as an early sketch, not a tested guarantee — worth knowing before relying on it.
MCP Server
MCP Server
For advanced setups: let another agent drive your entire Frontier install directly.
What an agent can do through the MCP server
Nearly everything you can click through in the app has an equivalent an outside agent can call directly.
Proxy: a browser your agents can use
Turn on the desktop app's own browser, and your agents can open pages, read them and act on them.
Proxy accounts and logins
One session per account, so agents work as the right person, and two accounts can run at once.
Teaching Proxy a site
Do a task once while Proxy watches, and it becomes a tool every agent can use.
Desktop & Remote Access
Opening Frontier from your phone
Pair your phone once, and the same Frontier, running on your machine, is there in your pocket.
Using Frontier away from home
Your phone reaches the computer at home from anywhere. You always know when it does, and what it changed.
Seeing what's actually running
One readout for the moving parts — both connections, the background loops, and whether your coding agent knows about Frontier.
Being told when something needs you
Agents can reach you outside the thread — sparingly, and only when waiting would cost you.
Asking for your approval
For the moments an agent shouldn't go on without a yes, it can ask and wait for your answer.
Letting agents use your phone
A photo, where you are, your clipboard or the share sheet, and your phone asks you every time.
Admin & Identity
Admin & User Profiles
For installs shared by more than one person — who's who, and who can manage what.
Sharing one install with your team
One Frontier install, running on one machine, can genuinely serve more than one person.
Moving a workspace between contexts
Handing off one project to someone else's space, without dragging your whole setup along with it.
Backing up and restoring your setup
Your workspaces, conversations, memories and settings can travel with you to a new machine.
Tutorials
Create your first workspace
The smallest complete loop — a repo, a workspace, and one real conversation.
Build a Watcher / Addresser strategy
Wire up two agents that hand work to each other — from a blank page to a running pipeline.
Set up an Engineer / Reviewer pipeline
A second set of eyes on everything you build — wired up once, running from then on.
Recover from a stuck agent
A quick checklist for the moment an agent seems to have gone quiet and you're not sure why.
Build an agent that runs on a script
A loop that does the mechanical part itself — no model run, nothing granted it didn't need.
Turn your data into a chart
Ask a question, get a picture of the answer under the reply, and click into it.
Teach Proxy a site in five minutes
Record a task once, save it as a capability, and put it behind a button.
Build a live dashboard for a workspace
An agent keeps the numbers current, and a table at the top of the workspace keeps up with them.
Find out where your tokens go
Ask which tools and workspaces spent the most this week, then fix the one that matters.
Still stuck?
Contact us