KAEONIX
SIGN INGET STARTED
POWERED BY AVATAR · ~12MB BINARY

Kaeonix Code

The agent that works where your code lives. Plan mode, OS-level sandboxing, subagents, session fork and undo — driven by Avatar, signed in with the Feyncode account you already have.

Install platform
brew install kaeonix/tap/kaeonix

Then kaeonix login — usage draws on your plan, no separate keys to manage.

kaeonix — ~/projects/api-gatewaysandbox: on · net: denied
add rate limiting to the /auth endpoints, then run the tests
avatar · thought 0.4s
Mapping the routes first. I stay in plan mode until the approach is approved, then apply edits and verify with the suite.
read · src/routes/auth.ts · 118 lines
grep · rateLimit · 0 matches
read · package.json · 76 lines
plan mode — mutating tools blocked while I map this out
1. add express-rate-limit to auth.ts — 100 req / 15 min per IP
2. extend auth.test.ts with a 429 case
approved · exiting plan mode
editing src/routes/auth.ts — LSP verified, 0 diagnostics
run npm test ?
auto-allowed — read-only test command, matches safe policy
48 passing (2 new) · 0 failing · 1.9s
done in 11s · /undo revert · /fork branch
CONTEXT 15%
LSP 0
MCP 2
01 / CONTROL

Autonomy you can audit.

01

Plan mode blocks at the runtime.

Research with every mutating tool refused by the runtime rather than discouraged by a prompt. Avatar reads, maps the change, and hands you numbered steps. Approve to execute, or redirect having spent a handful of tokens.

TOOL GATE
readALLOWED
grepALLOWED
editBLOCKED
bashBLOCKED
planAWAITING APPROVAL
02

Sandboxing the OS enforces.

Sessions run inside real isolation — bwrap on Linux, Seatbelt on macOS, Docker anywhere — with the network denied until you open it. A permission layer clears safe commands and holds risky ones for review.

ENFORCEMENT LAYERS
HOST FILESYSTEMREAD-ONLY OUTSIDE PROJECT
SANDBOX BOUNDARYBWRAP / SEATBELT / DOCKER
NETWORKDENIED
AGENT PROCESSPLAN MODE ACTIVE
03

The whole stack in one binary.

Plan mode, hooks, skills, subagents, fork and undo, LSP-aware self-correcting edits, command palette, transcript search. Existing MCP configuration is picked up on first run, so a switch takes seconds.

BINARY COMPOSITION · 12MB
AGENT CORE62%
TOOLS + LSP22%
SANDBOX10%
TUI6%
02 / TOOLKIT

Everything included.

Nothing to assemble, nothing held back for a higher tier.

AI permissions

Safe commands clear themselves; risky ones wait for you.

Subagents

Fan parallel work out to focused agents with clean contexts.

Hooks & skills

Shape behavior with lifecycle hooks and reusable skills.

LSP-aware edits

Edits read diagnostics and correct themselves in place.

Command palette

Jump anywhere and search full transcripts instantly.

Fork · undo · redo

Branch a session, roll edits back, replay with confidence.

MCP auto-import

Picks up your existing MCP configuration on first run.

Transcript search

Find any past turn, tool call, or diff by keyword.

Session ledger

Live context, spend, and diagnostics on every turn.

Editor bridge

Open any diff in your editor straight from the session.

Headless runs

Script the agent in CI with the same policies applied.

One sign-in

Your Feyncode account covers chat, terminal, and API.

INSTALL IN ONE LINE

Put the agent
where the code is.

brew install kaeonix/tap/kaeonix
CLI REFERENCEPRICING