https://github.com/geramy/nexus-project-client
Nexus Projects — source-available Application for AI-coordinated project & task management (text + voice Coordinator, plans, tasks, agents).
https://github.com/geramy/nexus-project-client
ai ai-agents ai-tools code harness programming project-management task-manager
Last synced: 7 days ago
JSON representation
Nexus Projects — source-available Application for AI-coordinated project & task management (text + voice Coordinator, plans, tasks, agents).
- Host: GitHub
- URL: https://github.com/geramy/nexus-project-client
- Owner: Geramy
- License: other
- Created: 2026-05-30T23:33:07.000Z (22 days ago)
- Default Branch: main
- Last Pushed: 2026-06-07T05:16:28.000Z (15 days ago)
- Last Synced: 2026-06-07T05:19:38.794Z (15 days ago)
- Topics: ai, ai-agents, ai-tools, code, harness, programming, project-management, task-manager
- Language: Dart
- Homepage: https://www.nexus-projects.ai/Home/Projects
- Size: 3.28 MB
- Stars: 3
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE.md
Awesome Lists containing this project
README
# Nexus Projects (Client)
A production-grade, multi-tenant Flutter client for managing complex projects with AI coordination, agentic task execution, Git integration, CI/CD, and observability.
## Vision
Nexus Projects gives teams (and individuals) a powerful environment where:
- **Clients** act as top-level tenants (full data isolation + sharing via "Clone to Client").
- **Projects** contain rich plans that drive hierarchical task generation.
- The **Project Coordinator** (the "main brain") can be talked to via **text or voice**, sees full project context, and can directly create/update tasks, propose plan changes, and generate diagrams using real multi-modal capabilities.
- Real **AEC + VAD** voice processing for natural, hands-free conversations.
- Full agentic workflows with inference servers, personas, tool calling, and live reactive updates.
## Architecture: Router + Client
Nexus Projects is two cooperating components — a server-side **Router** and this user-facing **Client**.
### Nexus Router (server)
A separate, multi-tenant gateway service (C#/.NET) that sits in front of all AI inference. It is the billing
and access-control brain of the system:
- **Authentication & tenancy** — registration, login, and API tokens, organized as a tenant chain of
**Client → Account → Users → Tokens** with strict data isolation.
- **Subscriptions & billing** — subscription plans and add-ons backed by Stripe, with checkout and
manage-billing flows.
- **Metering & quotas** — every routed inference request is measured against the account's entitlements for
the current Stripe billing period. Going over quota does not hard-fail requests; instead output is paced
(throttled) and an upgrade prompt is surfaced, so the app keeps working.
- **Inference proxy** — chat, tools, image, and audio (STT/TTS) requests are proxied through the Router to
the underlying inference backends, so usage can be authenticated and billed centrally.
The Router is closed-source and operated by Nexus Projects; this repository contains the Client only.
### Nexus Projects Client (this repo)
The Flutter desktop/mobile application — the project workspace, the AI **Coordinator**, plans, tasks, and
agents. The Client connects to the Nexus Router for **authentication, subscription/billing, and routed
inference**. The account surface in the Client (login, register, plan catalog, usage meter, checkout) is the
front end to the Router's subscription system; it is not a standalone billing implementation and is not meant
to be repointed at any other gateway.
In short: the **Router decides who may use inference and bills for it; the Client is where the work happens**
and is the customer-facing front end to that subscription system.
## Key Features (Current State)
- **Strict multi-tenancy** — Client selector scopes everything (projects, tasks, inference servers, personas, activity, etc.).
- **Project Coordinator** (the signature feature):
- Rich text chat with live project context.
- Voice mode with automatic end-of-speech detection (real Silero VAD).
- Proper OS-level AEC and noise suppression via `audio_session` + platform voice communication modes.
- The AI can call tools to **create tasks**, **update status**, **propose plan adjustments**, and **generate diagrams** — changes persist immediately and reflect in the UI.
- **One class per file** architecture with clean separation (`infrastructure/models/database` vs `ui`, `features/...`).
- Reactive Drift + Riverpod throughout.
- Supports streaming, tools, image generation, and audio (STT/TTS).
- 3-pane resizable shell with proper default proportions.
## Screenshots
#### Project setup, driven by an interview
The Coordinator interviews you about the project with short multiple-choice questions and proposes tags onto the board as you talk — turning a conversation into a structured stack (Industries, Platforms, Objectives, Features, Languages, Frameworks, Libraries). The left rail scopes everything by **Client → Project**.
#### The Coordinator interview — text or voice
Answer by typing, or go hands-free. Real Silero VAD with end-of-speech detection and OS-level echo cancellation lets you talk to the Coordinator naturally while it fills in the board live.
#### Living plans
Generated plans are living Markdown documents (Overview, Client, Server, Database) that every layer shares as project-wide context. Choose the Coordinator agent that owns each plan.
#### Hierarchical tasks with a rich detail panel
Each task has Overview, Sub-Tasks, Git Changes, Agent Work, and Builds & CI tabs. Status, priority, dates, and assigned agent persist live.
#### Agents & personas
A roster of specialized agent personas — Project Manager, Coordinator, and SDE specialists (Generalist, Networking, Physics, Database, UI/UX, DevOps) plus a Verification agent — each backed by a configurable model.
#### Bring your own inference
Automatic LAN discovery of Lemonade inference servers alongside your configured providers — pick the model that runs each agent.
#### Code & Git, in app
An integrated workspace with Git — commit, push, pull, merge, branch, and history — plus an editor and source-control view over your project's plans and code.
#### Build & package from the app
Docker-backed build environments, images, and containers with live build logs.
#### Account, usage & subscription
Usage is metered against your plan for the billing period; going over quota **paces** output instead of hard-failing. Plans run from Basic to Platinum with optional add-ons.
## Project Structure
```
nexus_projects_client/
├── lib/
│ ├── features/
│ │ ├── projects/ # Coordinator chat + session (core voice work)
│ │ ├── project_plans/
│ │ ├── tasks/
│ │ └── ...
│ ├── infrastructure/
│ │ ├── inference/ # InferenceClient (chat, tools, image, audio)
│ │ ├── database/
│ │ └── models/
│ └── services/audio/ # AEC, VAD, recorder, TTS
├── packaging/ # Windows Inno Setup installer script
├── .github/workflows/ # CI + desktop build + release pipelines
└── README.md
```
## Getting Started
```bash
flutter pub get
dart run build_runner build --delete-conflicting-outputs
flutter run -d macos # or your preferred device
```
### Voice / Microphone
On first use you will be prompted for microphone permission. The app uses proper voice communication audio modes for echo cancellation when talking to the Coordinator.
## Technology
- Flutter + Riverpod + Freezed + Drift (reactive SQLite)
- GoRouter
- Dio (with full streaming + tool call assembly)
- `record` + `just_audio` + `audio_session` + `vad` (for high-quality voice)
## Contributing / Philosophy
We follow extremely strict organization rules:
- One class per file
- Clear separation between database models and UI models
- All new major features (especially anything touching the Coordinator) must live in their own well-named files under `features/` or `services/`
## License & Usage
The Nexus Projects Client is **source-available** under the Sustainable Use License with Additional Use
Restrictions — see [`LICENSE.md`](LICENSE.md) for the full terms. It is source-available, not OSI "open
source." The following restrictions are central:
- **The subscription system on the Client may not be modified or replaced — at all.** You may not alter,
reconfigure, bypass, or substitute your own subscription, billing, entitlement, or metering logic, and you
may not point the Client at any inference/billing gateway other than the official **Nexus Router**.
- **The only permitted alternative is to remove the subscription system entirely** — i.e. ship a build that
contains no subscription, billing, or entitlement functionality whatsoever. There is no middle ground:
either keep the official subscription system and Router unchanged, or have none of it. Building your own in
its place is not allowed.
- The **login / registration (authentication)** system likewise may not be modified to authenticate against
any service other than the official Nexus services.
### Commercial use of the Router / subscription system
If you want to use the Nexus Router or its subscription/billing routing **commercially**, that is available
under a separate commercial license — please get in touch at **support@nexus-projects.ai** to arrange it.
**Organization exceptions:** **Lemonade-sdk** is permitted to use the Router through an existing partnership.
No other organization has this exception except by separate written agreement.
---
Built iteratively with a strong focus on real agentic workflows and natural voice interaction with your project's "main brain".