Project status

Capability self-check, current limitations, release cadence and tech stack — what is done, what is not, and every number links back to source.

Architecture health trend (9 snapshots) will land on this page in the next release.

Capability self-check

Sixteen dimensions, each with its current status and its source location — every number links to code, and what is not finished yet is written down too.

Dimension	Liri	Status	Source
Language	TypeScript + Rust	Shipped	View
Local-first	Ollama / llama.cpp + local SQLite	Shipped	View
Smart routing	19 task types · 4 tiers · 5 decision layers	Shipped	View
Security model	5 defense layers · 6 sandbox backends	Pending audit	View
Messaging channels	27 channels (incl. WeChat / DingTalk / Feishu / QQ)	Shipped	View
Desktop client	Tauri 2.11 · 429 TSX components	Polishing	View
Terminal UI	React + Ink	Shipped	View
Persona system	18 species × 8 hats × 6 eyes	Shipped	View
Voice interaction	Local faster-whisper / Piper + cloud	Shipped	View
Cost transparency	Pricing in the DB · every call recorded	Shipped	View
Second Brain	Knowledge base + memory + event log	Shipped	View
Context governance	Layered windowing + three-stage compression	Shipped	View
Marketplace	Plugin market + MCP market + skill sources	In progress	View
Agent strategies	4 strategies · PDCA · Swarm	Shipped	View
Scheduled tasks	once / interval / cron	Shipped	View
Protocol support	MCP + ACP + A2A + LSP	Experimental	View

Status is self-assessed by the maintainers and moves with each release; “View” jumps to the relevant directory so you can check it yourself.

* Data reflects source v0.5.22; source links point to the main branch on GitHub.

Current limitations

What is not finished yet during rapid iteration — written down so you do not have to discover it the hard way.

Security awaits external audit

Pending audit

5 defense layers and 6 sandbox backends are implemented, with 51 sandbox-related files you can review one by one — but no third-party security audit has been completed yet.

Desktop interactions still being polished

Polishing

429 TSX components and 32 pages are in place; shortcuts, animations and edge cases are still being refined.

Marketplace catalog is still small

In progress

The plugin market, MCP market and skill sources work end to end, but the installable catalog is still growing.

ACP / A2A are experimental

Experimental

Both protocols are still evolving and their interfaces may change between releases; MCP and LSP are stable.

English docs may lag behind

In progress

Chinese docs are synced automatically from the source repository; the English version is a manual translation and is usually one release behind.

The repository wins over the site

Note

These pages are generated from app/docs. If something disagrees with the source, the GitHub repository is authoritative.

Found an issue? Open an Issue

Versions & Cadence

Version numbers, tag counts and commit history are all verifiable on GitHub; the history includes pre-launch development, with public iteration starting at launch

2026-10-28 v0.5.22 Current release

69 built-in tools across 23 categories
27 messaging channels
5 security layers + 6 sandbox backends
429 desktop components · 32 pages

2026-10 v0.5.13 → v0.5.21 Shipped

Eight releases in three days (v0.5.13 → v0.5.21), all part of pre-launch development
Core modules extracted, guardrail baselines synced
Progress tracking now driven by artifact existence
Sandbox permission changes apply instantly

Repository totals 118 tags Release rhythm

3,325 commits to date
CI covers lint:arch / type checking / unit tests
Public repository with visible issues and PRs

Next v0.6+ Planned

Tracked in GitHub Releases and Issues
Online documentation site is live (Chinese and English)
Vote on feature priorities in Issues

Tech stack

A modern stack, chosen with care at every layer

TypeScript 4,109 files Core language · ~870K lines

Rust 9 modules Native modules for AST, security and compression

Bun Runtime

React 19 + Ink Terminal UI

Tauri 2.11 429 components Desktop client

SQLite Data and vector storage

OpenTelemetry Observability

MCP + ACP + A2A + LSP Protocols

Docker One of the sandbox backends

MIT License Open-source license

Local-first

Run local models through Ollama or llama.cpp, and keep your data on your own machine by default. When you switch to a cloud provider, only the contents of that request are sent to the provider you chose

🏠

Works offline

With a local Ollama or llama.cpp model you can chat and run local tools without a network connection. Sessions, memory and the knowledge base all live in local files.

🔒

Your data stays local

Sessions, memory and the knowledge base live in local SQLite and your workspace directory, with conversation history appended to events.jsonl for easy review and auditing.

🌐

Add the cloud on demand

When you need more horsepower, enable cloud providers (14 supported) in your config and let routing pick one per task tier — or stay local only.

Local vs. Cloud

Where the model runs

Your machine (Ollama / llama.cpp) The provider’s servers

Network required

No Yes

Where request data goes

Never leaves your machine Sent to the chosen provider

Cost model

Your own hardware Pay per use

Conversation history

events.jsonl stays local Stays local too

Updated with each release; the GitHub repository is authoritative if anything disagrees.