Aurict CLI
Open-source terminal AI coding agent. Bring your own model, keep context local, approve every action.
Explore Aurict CLI
one company · connected products
The Aurict ecosystem
One company. Connected products. Built around you.
The company. The connection.
Hoprel
Local-first desktop AI agent workspace for macOS, Windows and Linux.
- Read the project files
- Plan the change
- Waiting for your approvalApprove
Aurict Mobile
Our studio of consumer apps and games.
Jelly Paradox · Duckley · Mind Grid
Meet our studio
Aurict Decision
Private, on-device decisions in under 10 ms.
Raw data never leaves the device.
Explore Aurict Decision
works withAnthropicOpenAIGooglexAIAzureAWS BedrockOllamaOpenRouter+4 more — bring your own key
why aurict
Designed to understand the work before it answers the prompt.
Aurict does not replace one chat window with another. It makes your project, risk, and the shape of the work visible.
Agent-first, not prompt-first.
Every task routes to a domain specialist — explore, code, review, test, debug, docs, security, performance, analytics — each with its own tools and context budget. Complex work decomposes and runs in parallel instead of waiting in a single queue.
Reads your codebase before your prompt.
On first run, Aurict walks your directory tree, detects framework and package manager, and ranks the files most likely to matter — no manual context to attach, nothing to paste.
Zero-setup, genuinely cross-platform.
One install command ships a native compiled binary for macOS, Linux, and Windows — no Node runtime at execution time, no WSL detour, no Rosetta tax on Apple Silicon.
Open, and built to be extended.
Bring compatible MCP servers, custom commands, and shared team skills. Imported servers are visible and inspectable in the terminal — nothing is locked behind a proprietary cloud.
capability architecture
Understand the work. Coordinate it. Act safely.
Every capability is part of the same operating path from context to result.
bash classifier
Every shell command is analyzed before it runs — safe commands execute instantly, destructive ones stop for confirmation.
sandbox execution
Risky calls pass through a policy sandbox: approvals, protected paths, scrubbed env, timeouts, and an audit trail.
multi-agent orchestration
Nine specialists in isolated workers — explore, code, review, test, docs, security, debug, performance, analytics.
218+ contextual skills
Framework and tool-specific context, injected automatically the moment your stack is detected.
mcp client
Import compatible existing configuration, then inspect each connected server and its tools before relying on it.
windows native
A real compiled binary for x64 — no WSL, no PowerShell workarounds. Shell auto-detected.
persistent memory
Decisions and conventions stored locally and recalled per-session — across every restart.
design agent
150+ design systems and 111 skill templates — describe a UI idea, pick a system, get a full implementation brief.
verified completion
Changed files, verification evidence, open work, and waivers remain visible in a durable completion proof.
workspace intelligence
Search source semantically, inspect installed package APIs, read workspace images, and verify work through browser or eval tools.
security posture
Confirmation is the default, not an afterthought.
Security capability starts hidden from the model entirely. You choose how far Aurict is allowed to reach.
Passive
Defensive review and reporting only. No offensive tooling is exposed to the model.
Active Lite
Controlled, Docker-backed scans against an allowlisted target — rate-limited and permissioned.
Kali Full
The larger experimental image, for when you deliberately need the full toolset. Approval required every time.
$ aurict /config security allow <target> — every active profile still requires an allowlisted target and explicit approval.
install
Thirty seconds, one command, your existing shell.
macOS, Linux, or Windows — same command everywhere.
Launch inside any project directory.
# provider, key, model, Project AutoInteractive setup — bounded project edits can be session-approved; sensitive work still asks.
Optional. Targets still need explicit allow-listing.
open ecosystem
Works with the tools you already run.
via MCP — connect anything with a config file
GitHubPostgreSQLDockerSlackJiraLinearFigmaSentryVercelAWSNotionBrowser
start in your own context
Start where your work belongs.
faq
Questions engineers actually ask.
The short answers you need to make a decision.
Yes — Aurict is AGPLv3 licensed and open source on GitHub, free for individuals and teams alike. You bring your own model provider key.