$ agent-interface --with agent-backbone

Agent Backbone, right beside your terminals.

Agent Interface runs any terminal program on its own. Connect Agent Backbone and your agents' live states, team, activity, usage, skills and policies appear as native UI next to the sessions they run in.

storefront 5 tabs
Jump to tab, pane, agent ⌘K
TABS
1 api
2 web
3 db
4 docs
5 release
AGENTS 3 need you
reviewer needs you api
planner needs you release
deployer needs you deploy
builder working api
web-dev working web
docs-writer idle docs
builder working 1
> Add rate limiting to the public API, then push
it for review.
⏺ Search(pattern: "Router::new", path: "src")
⎿ Found 2 files
⏺ Read(src/routes/mod.rs)
⎿ Read 142 lines
⏺ I'll add a token-bucket limiter as middleware
and apply it to the /v1 routes only.
⏺ Write(src/middleware/limit.rs)
⎿ Wrote 64 lines
⏺ Update(src/routes/mod.rs)
⎿ Updated with 2 additions
18 + .layer(RateLimit::per_minute(120))
19 + .route_layer(limit::by_client())
⏺ Bash(cargo test --all)
⎿ test result: ok. 48 passed; 0 failed
⏺ Bash(git push -u origin feat/rate-limit)
⎿ branch feat/rate-limit set up to track origin
✶ Writing tests for burst traffic… (41s · esc to interrupt)
>
? for shortcuts
reviewer needs you 2
› Review feat/rate-limit from builder.
• Ran git diff main...feat/rate-limit --stat
└ 2 files changed, 66 insertions(+)
• Ran cargo clippy --all-targets
└ Finished with no warnings
The limiter looks right. One concern: clients
are keyed on X-Forwarded-For, which can be
spoofed behind a proxy. Key on the API token?
› 1. Yes, write it up as a review note
2. No, approve as is
›
Skills every skill an agent can load, by where it lives
Shared skills One copy in the store; Backbone links each to the agents whose tags it names.
allevery agent
commit-style
review-checklist
rust3 agents
cargo-workflowalso review
api-security2 by name
clickread the skill a repository skill wins over a store skill of the same name
storefront api pane 1 of 3 ⌘Dsplit ⌘⇧Dsplit below ⌘⇧↩expand ⌘Kjump agents shown Saved
storefront 5 tabs
Jump to tab, pane, agent ⌘K
TABS
1 api
2 web
3 db
4 docs
5 release
AGENTS 3 need you
reviewer needs you api
planner needs you release
deployer needs you deploy
builder working api
web-dev working web
docs-writer idle docs
builder working 1
> Add rate limiting to the public API, then push
it for review.
⏺ Search(pattern: "Router::new", path: "src")
⎿ Found 2 files
⏺ Read(src/routes/mod.rs)
⎿ Read 142 lines
⏺ I'll add a token-bucket limiter as middleware
and apply it to the /v1 routes only.
⏺ Write(src/middleware/limit.rs)
⎿ Wrote 64 lines
⏺ Update(src/routes/mod.rs)
⎿ Updated with 2 additions
18 + .layer(RateLimit::per_minute(120))
19 + .route_layer(limit::by_client())
⏺ Bash(cargo test --all)
⎿ test result: ok. 48 passed; 0 failed
⏺ Bash(git push -u origin feat/rate-limit)
⎿ branch feat/rate-limit set up to track origin
✶ Writing tests for burst traffic… (41s · esc to interrupt)
>
? for shortcuts
reviewer needs you 2
› Review feat/rate-limit from builder.
• Ran git diff main...feat/rate-limit --stat
└ 2 files changed, 66 insertions(+)
• Ran cargo clippy --all-targets
└ Finished with no warnings
The limiter looks right. One concern: clients
are keyed on X-Forwarded-For, which can be
spoofed behind a proxy. Key on the API token?
› 1. Yes, write it up as a review note
2. No, approve as is
›
Skills every skill an agent can load, by where it lives
Shared skills One copy in the store; Backbone links each to the agents whose tags it names.
allevery agent
commit-style
review-checklist
rust3 agents
cargo-workflowalso review
api-security2 by name
clickread the skill a repository skill wins over a store skill of the same name
storefront api pane 1 of 3 ⌘Dsplit ⌘⇧Dsplit below ⌘⇧↩expand ⌘Kjump agents shown Saved

with Agent Backbone connected: the agents section, pane states and the Skills view

what it adds

Every agent in view, without leaving the terminal.

AGENTS1 needs you
reviewerneeds youapi
builderworkingapi
docs-writeridledocs
sidebar · agents

Your agents, in the sidebar

Every agent with its state and the tab it runs in, the ones that need you first. Click one to land on its session.

reviewerneeds you2
Key on the API token instead?
› 1. Yes, write it up
2. No, approve as is
panes · states

Panes that ask for you

A pane whose agent is waiting on you gets a ring and a clear label, so a decision never sits unnoticed.

Team6 agents · 5 onlineFolderTagSwarm
! 1 needs you2 active2 idle1 offline
├builderclauderust · apiworking
└reviewercodexreviewneeds you
view · team

The whole team at a glance

Agents grouped by folder, tag or swarm, with the CLI each one runs and what it is doing now.

09:0009:30now
builder
reviewer
planner
workingneeds youidle
view · activity

The last hour, lane by lane

A timeline of what each agent did and when it waited for you, drawn from the states Backbone observed.

Tokens today[value]
cache readscache writesoutput
view · usage

Usage, per agent and overall

See where tokens go, from cache reads and writes to output, for one agent or the whole team.

Skillsevery skill an agent can load
rust3 agents
cargo-workflowalso review
api-security2 by name
commit-styleall
views · skills · policies · tags

What every agent loads

Skills, policies and tags, and how each reaches your agents. Open any skill to read it right in the pane.

how it connects

Read only. Local only. Honest about what it knows.

the window
Agent Interface

Shows states and views next to the sessions they belong to.

reads over⟵local HTTP
the coordinator
Agent Backbone

Knows your agents: their states, skills, policies, tags and usage.

tracks⟶agents
your work
Your agents
claudecodexany CLI
read only

Agent Interface only reads. Nothing in the integration approves, tags or edits your agents.

local only

Requests go to an approved endpoint on your own machine, each with a three-second deadline.

never stale

If a refresh fails, the last state is marked stale. It is never shown as current.

always fresh

The agents list refreshes two seconds after each read, Team every three seconds and Usage every fifteen.

set up

Three steps to connect.

01
Run Agent Backbone
Install and start it on the same computer, following its own guide.
agent backbone guide ↗
02
Share its port and key
Set these in Agent Interface's environment, or in the .env file of its library folder.
BACKBONE_PORT=[port]
BACKBONE_API_KEY=[your key]
03
Open Agent Interface
The agents section appears in the sidebar, and every empty pane offers Skills, usage and team. Hide it any time from the status row.

optional · Agent Interface is a complete terminal workspace without it

questions

Agent Backbone, briefly.

What is Agent Backbone?

A separate open-source project that keeps track of your agents: who they are, what state they are in, and which skills, policies and tags reach them. Agent Interface is its window.

Do I need it to use Agent Interface?

No. Workspaces, tabs, split panes, Jump and tmux sessions all work without it. The agents section and views appear only when Agent Backbone is connected.

Can Agent Interface change my agents?

No. The integration is read only: it shows what Agent Backbone knows and never approves, tags or edits anything.

Which agents does it work with?

The ones Agent Backbone tracks, whichever CLI they run, such as Claude Code or Codex. The Team view shows each agent's CLI beside its name.

Can you help us set up and build with Agent Backbone?

Yes. Our support for teams and individuals covers Agent Backbone setups, complex configurations and solutions built with it.

Read more

all posts →
how it works · [n] min read
Close the window, keep the work: how Agent Interface uses tmux
integration · coming soon
What Agent Interface reads from Agent Backbone, and why it never guesses