Skip to content
NeryaAIPublic

About

Skill-first, trading-native, self-evolving autonomous agent runtime for AgentNetwork workflows.

Topics

Resources

Stars

15 stars

Watchers

0 watching

Forks

Latest commit

 

History

124 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Nerya

Nerya

One idea. A team to build it.

Your local Agent strategy workspace — research, build, backtest and review in one conversation.

1.0.0 Beta · 1.0.0-beta.1

English · 简体中文

Website · Documentation · Desktop downloads · Build status

Desktop CI Version License

Describe your trading idea and its constraints. Nerya brings researchers, strategy authors and reviewers into the same workspace. Follow the tools as they run, inspect the evidence, open the strategy and backtest details, and decide what happens next.

Nerya strategy conversation and workflow, English

Recorded from the website's isolated product tour. It uses example data, not a live account or verified investment performance. Recording provenance.

From idea to a traceable workflow

Build a strategy. Describe the market, entry and exit logic, time horizon and constraints. Agents prepare the strategy package and validate it. Script strategies, script-driven Agent strategies and event-driven Agent strategies share the workspace, with readable workflow nodes, parameters and run history.

Research as a team. Delegate market data, news and technical research to specialized Agents. A reviewer can challenge assumptions while the lead combines the evidence. Open source material and research reports without losing the conversation.

Inspect more than a text reply. Strategy cards, backtest curves, market charts and research reports are part of the conversation. Open their details in workspace tabs. Backtest views include available candle history, entry/exit markers, protection levels, trades and execution evidence; missing data remains visible rather than being presented as a complete result.

Review and iterate. Inspect a particular run and its logs before changing a strategy. Optional review schedules produce evidence-linked recommendations and candidate changes. Review and trading authorization are distinct from editing ordinary Skill or Agent resources: those resources can be edited and saved directly.

Keep the workspace local. Strategies, cached market history, reports and configuration live in your workspace. Credentials are stored encrypted in the local Vault and referenced by identifier. Configured model providers, market services and external tools still receive the requests needed to perform your tasks; local storage does not make those services offline.

What is in 1.0.0 Beta?

Area In the workspace
Agent conversation Streaming replies, available thinking output, tool execution, progress, follow-up messages and recoverable errors
Strategies Script and Agent execution, timer/event workflow nodes, strategy editing, import/export and run details
Backtesting Reusable local historical data, preflight checks, coverage information, per-market charts and trade records
Research Market cards, technical charts, evidence-linked reports and multi-Agent research
Skills & Agents SKILL.md playbooks, lazy-loaded references, reusable scripts and direct resource editing
Integrations Configurable model providers, exchange/wallet connections, MCP and external Agent access
Desktop Tauri host with bundled Python, Node, the web workspace and Chromium; native build pipelines for three operating systems
Interface English/Chinese, light/dark themes, workspace tabs and settings-based account/model configuration

Research categories and execution support are not interchangeable. Crypto exchanges use the configured connectors; wallet and prediction-market operations depend on the relevant provider and credentials. Futures and A-share research examples do not imply that every broker, order type or instrument is executable.

Install the desktop app

Choose a completed release on GitHub Releases, then download the installer matching your operating system and CPU.

Platform Architecture Release files
macOS Apple Silicon / ARM64 .dmg, .pkg
macOS Intel / x64 .dmg, .pkg
Windows x64 NSIS .exe, Windows Installer .msi
Linux x64 .deb, .AppImage

Desktop installers contain the application's Python and Node runtimes, web server and browser engine. You do not need to install a development toolchain to use them. Windows installers include the offline WebView2 installer. Linux still depends on distribution-provided desktop/browser libraries; the build baseline is Ubuntu 22.04. See desktop installation and build notes.

The beta pipeline uses an ad-hoc macOS signature and unsigned Windows packages unless a separate signing process is supplied. These are not notarized Apple Developer ID releases or Windows reputation-approved installers. Follow your organization's security policy; do not disable OS security protections to install an untrusted build.

On first launch, configure a model, add an account when needed, and configure the administrator password for external access. New workspaces contain no default strategies or scheduled trades. The account, model and access settings remain available afterward in Settings. Upgrades do not silently erase an existing workspace.

Try a first task

Research BTC/USDT using the available historical data. Build a script strategy with explicit entry and exit rules, validate it, and run a historical backtest. Show the strategy workflow, data coverage and trade details. Do not start live trading.

Or ask for a research-only result:

Compare BTC and ETH market conditions. Split the research between price action and news, cite the evidence, and produce a report with market charts. Explain any missing data.

Model access and market-data availability depend on your configured services. A backtest is historical simulation, not a promise of future performance. Inspect the strategy code, parameters, coverage, fees and trading permissions before execution.

Run from source

The source workflow is for developers and server deployments. Use the Node and Python versions pinned in .node-version and .python-version.

git clone https://github.com/NeryaAI/Nerya.git
cd Nerya
uv venv
uv pip install -e ".[dev,mcp,trading,prediction,browser]"
npm ci
npm --prefix desktop ci

For desktop development:

npm --prefix desktop run dev

The desktop launcher manages its local services. A source build additionally requires Rust and the platform's native build dependencies. For a web-only deployment, use the existing CLI and MCP guide and nerya --help; the web dashboard remains independent of Tauri.

Build, verify and release

python scripts/release_version.py
npm run typecheck --workspace @nerya/dashboard
npm run check:i18n --workspace @nerya/dashboard
python -m unittest discover -s desktop/tests -v
npm --prefix desktop run build
python desktop/scripts/verify_runtime.py
python desktop/scripts/smoke.py --access-port 0

Run Python commands in the project virtual environment. macOS additionally provides node desktop/scripts/installer.mjs after building the app. The desktop guide contains exact native bundle commands and installer-payload checks.

The desktop workflow builds on Windows, Linux, Apple Silicon macOS and Intel macOS separately. It checks version consistency, locked dependencies, frontend types, Python distribution contents, desktop unit tests and the packaged runtime. Installer payload smoke tests use temporary workspaces and fake credentials, without calling a model or placing a trade. A build is not considered verified merely because the Rust compiler exits successfully.

Push a tag matching VERSION (for example v1.0.0-beta.1) to request a release. Only after the required workflow jobs succeed does the release job publish the installers, SHA-256 checksums and build metadata. Normal pushes and pull requests build artifacts but do not publish a release. Native notification prompts, OS trust dialogs and broker-specific live execution still require their own acceptance testing.

Project layout

nerya/              Python Agent runtime, APIs, connectors and trading services
nerya/skills/       Built-in SKILL.md playbooks, references and executable helpers
nerya/sdk/          Runtime strategy, trading and trigger APIs
sdk/               Python and TypeScript client SDKs
dashboard/         Next.js conversation and strategy workspace
desktop/           Tauri shell, portable runtime packaging and native smoke tests
tests/             Runtime regression tests
.github/workflows/ Validation, native desktop builds and release publication

Security and licensing

Live execution requires the corresponding configuration and authorization; do not bypass the trading risk and approval gates. Vault encryption protects stored credentials, not a compromised running process. Keep backups of your workspace and its encryption key, and never commit account secrets, .env files or live trading state.

Nerya is licensed under PolyForm Noncommercial 1.0.0. Commercial use requires a separate license. Bundled third-party components retain their own licenses. This beta is software for research and strategy operations, not investment advice or a guarantee against financial loss.

About

Skill-first, trading-native, self-evolving autonomous agent runtime for AgentNetwork workflows.

Topics

Resources

Stars

15 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages