# SyrinFlow

**Repository Path**: xqshop/syrin-flow

## Basic Information

- **Project Name**: SyrinFlow
- **Description**: 多智能体协同
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-09-05
- **Last Updated**: 2026-09-13

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# SyrinFlow

SyrinFlow is a focused multi-agent workspace for market research.
It uses Syrin for agent collaboration while keeping data, skills, tools, runs, events and reports
in an application-owned persistence layer. Market-data and quantitative adapters are optional
capabilities that can be added to a workflow when a research question needs them.

## Current Scope

- Create and track research runs from a Vue management console.
- Add multiple versioned Skills and retrieve them by name, description, body or tags.
- Define Agents, controlled tools and dependency-based workflows from the management console.
- Register RSS/Atom or GDELT sources, deduplicate articles and persist source provenance.
- Register public-page browser sources with Playwright selectors; challenge pages are recorded as failures and never bypassed.
- Persist research plans, agent events, reports and data-snapshot metadata.
- Collect news, market and other approved data into normalized, checksummed snapshots.
- Configure macro indicators from the console; World Bank collection is available without a key,
  while FRED/AKShare and paid providers use the same adapter contract and environment-secret references.
- Keep market-data, backtesting and notification providers behind replaceable adapters.
- Develop against SQLite locally and switch to the dedicated PostgreSQL database for sharing.

## Repository Layout

```text
apps/web/            Vue 3 management console
src/syrinflow/       FastAPI application and persistence models
skills/              Source-controlled built-in Skill documents
scripts/             Database and development utilities
docs/                Architecture and integration plans
vendor/syrin/        Pinned Syrin source submodule
```

## Start

```powershell
cd D:\SyrinFlow
Copy-Item .env.example .env
.\.venv\Scripts\python.exe -m uvicorn syrinflow.main:app --host 127.0.0.1 --port 8010
```

In a second terminal:

```powershell
cd D:\SyrinFlow\apps\web
npm run dev -- --host 127.0.0.1 --port 5174
```

Open `http://127.0.0.1:5174`.

For optional browser collection and scheduling dependencies:

```powershell
.\.venv\Scripts\python.exe -m pip install -e ".[collection,scheduling]"
.\.venv\Scripts\playwright.exe install chromium
```

Set `SYRINFLOW_SCHEDULER_ENABLED=true` in one dedicated process to enable the cron expressions
stored in each source's configuration. Keep it off in multiple API workers to avoid duplicate
polling.

## Syrin native Playground

The management UI above is part of SyrinFlow, not Syrin itself. To compare it
with Syrin's built-in multi-agent chat and debugging Playground, run:

```powershell
.\.venv\Scripts\python.exe .\scripts\serve_syrin_playground.py
```

Open `http://127.0.0.1:8020/playground`. The demo uses Syrin's local mock model,
so it demonstrates the native UI and routing surface rather than real analysis.

The default configuration uses a local SQLite file so the application can boot before shared
PostgreSQL credentials are supplied. For shared persistence, set `SYRINFLOW_DATABASE_URL` in
`.env`, then run `python scripts/create_database.py` with the administrator URL and dedicated
application-role password configured.

## Status

The configured market-research workflow is compiled into Syrin's native `Workflow` and executed
by Syrin's `WorkflowExecutor`; its lifecycle hooks and handoffs are persisted by SyrinFlow. The
generic `/api/runs` path still demonstrates a persisted Syrin `Swarm` for snapshot-oriented
research, with deterministic Mock models by default. Real LLM profiles, broader quantitative
tools and notification delivery are extension slices rather than hidden assumptions.

The default market-research flow is staged: collection, persistence, analysis, evidence review,
then report persistence. See `docs/AGENT_ORCHESTRATION.md` and the native graph endpoint
`GET /api/orchestration/workflows/{workflow_id}/graph`.

The current-affairs template is configurable and runs collection, database retrieval, analysis,
evidence review and report publishing as persisted workflow nodes. A failed news source is recorded
without stopping healthy sources. Agent responses still use deterministic Mock models until a real
LLM profile is configured.

Macro endpoints are available under `/api/macro`: indicator definitions, collection, latest AI Context
and macro snapshot listing. Standardized observations are stored under `data/snapshots/macro`.

## Migration

Current implementation status is recorded in `docs/PROJECT_STATUS.md`. Windows setup, secret handling,
database backup/restore and second-computer verification are documented in
`docs/TRANSFER_GUIDE.md`. Raw news, reports, database archives and `.env` are intentionally excluded
from Git.