Skip to content

markl-a/phantom-flow

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

phantom-flow

CI

Self-hosted + cluster-aware + AI-native + cross-device workflow engine on top of phantom-mesh — n8n / Zapier 的本地化替代品,招聘對齊 鴻海 C3 / Modal / 中型 AI 新創,副業 NT$199-499/月 自架 Pro tier。

status: alpha · Tier 1 license: Apache-2.0 phantom-mesh ecosystem

30-second demo

docs/demo.cast — asciinema recording of phantom_flow.runner flows/jobseek-daily.yaml --dry-run parsing a cron-triggered 6-step pipeline.

# play in a terminal (requires asciinema)
asciinema play docs/demo.cast

# or view the captured text without any tooling:
cat docs/demo.cast | jq -r '.[] | select(.[1]=="o") | .[2]'

Self-hosted on purpose — no upload to asciinema.org, no third-party tracking.

一句話 niche

n8n / Zapier / Make 都是 cloud-first + 不會 cluster dispatch;Temporal 是 重量級 JVM;Apple Shortcuts 是 Apple-only;LangChain 不是 event-driven。 phantom-flow 是第一個 self-hosted + cluster-aware + AI-native + cross-device workflow engine — 同一份 YAML 可以 dispatch 到你的 GPU 機、always-on Pi、或 手機,Rust+Python 原生 + 100% local-first。

Competitor Their edge Why phantom-flow exists
n8n 600+ integrations, visual Needs Node.js; not cluster-aware; not AI-native
Zapier 7,000+ apps SaaS only — phantom is on-prem / local-first
Make Polished visual editor Not cluster-aware; cannot dispatch to a GPU node
Temporal.io Enterprise-grade workflow Heavy JVM; phantom-flow is lightweight Rust+Python
LangChain LLM chaining Not event-driven; phantom-flow is
Apple Shortcuts Simple Apple-only; phantom-flow runs on 5 OSes

Status (2026-05-22)

  • Tier 1 shipped:
    • phantom_flow/runner.py — YAML flow executor (n8n-style) with --dry-run planner
    • phantom_flow/llm_driver.py — swaps LangChain LLM → phantom event capture for one code path
    • flows/jobseek-daily.yamlreal daily-running flow (cron 09:00)
    • flows/example-webhook.yaml — declaration-only spec demo
    • 兩個既有 repo subtree-merged: ai_automation_framework/ (17+ automation tools, RAG, agent framework, persistent memory) + data_analysis/ (Streamlit app, K-Means/DBSCAN/RFM/CLV, 377 pytest, multi-LLM cost routing)
  • 🟡 Tier 2 next: 全面 LLM provider 改走 phantom-mesh provider trait (現在只 1 條 code path);memory backend SQLite/Redis → phantom FTS5; 3+ 個真實 flow(現只 1 個 jobseek-daily)。
  • 🟡 Tier 3 (M2-M3, ~2026-07): visual flow editor、marketplace templates、 健康 + 學習 flow。

30-second quickstart

git clone https://github.com/markl-a/phantom-flow
cd phantom-flow

# 跑 jobseek-daily flow 的 dry-run(印 trigger + pipeline + outbound,不碰
# 網路 / 檔案)
python -m phantom_flow.runner flows/jobseek-daily.yaml --dry-run

# 跑真實版(需 merged subtree code + Python deps)
python -m phantom_flow.runner flows/jobseek-daily.yaml

Architecture (within phantom-mesh ecosystem)

phantom-flow 是 P2 分身連線 + P3 進化網 的執行層:把 phantom-mesh 的 event capture / FTS5 memory / provider trait 串成可重複的 workflow,並用 cluster-aware dispatch 把重負載送到對的 mesh 節點。

YAML flow (flows/*.yaml)
   ↓
phantom_flow.runner  (n8n-style executor + dry-run planner)
   ↓
phantom_flow.llm_driver  →  phantom-mesh provider trait
   ↓                         (Claude / Gemini / local / Modal spill)
trigger (cron / webhook / event)
   ↓
dispatch decision  →  Mac M-series  |  GPU node  |  always-on Pi  |  phone
   ↓
phantom event capture  →  FTS5 memory  →  phantom-companion ⑦ 觀察

Pillars served: P2 (分身連線 — workflow 跨裝置 dispatch)、P3 (進化網 — 每次 flow 執行回寫 FTS5,讓 ⑦ companion 學模式)、P1 (跨平台 — 5 OSes)。

Target users (recruiter / co-builder angle)

  • Recruiters (招聘): 鴻海 C3 AI Service / Modal / Together / 中型 AI 新創。Workflow engine 經驗在 2026 是 high-signal: 證明能做 distributed systems + DAG scheduler + LLM cost routing + 真實 production cron。Rust + Python 雙語、AI-native 是 differentiator。
  • 副業 angle (NT$ 199-499 / 月 Pro tier): n8n self-hosted 市場已被驗 證,中文圈 + AI-native + 跨裝置這個位置目前是空的。+ Hahow 課程「自架 phantom-flow 取代 Zapier」+ premium template marketplace。
  • 人生 (life): 直接吃 C4 (注意力切換)、C5 (求職 + 副業)、C7 (個人 財務)、C8 (家人健康) — 重複的事 automate 掉。
  • Co-builders: 想要 self-hosted + AI-native 工作流的 indie hacker / homelab 玩家;跨裝置 dispatch 是其他 OSS 沒有的差異化。

Roadmap (per master plan)

3-bullet:

  1. M2 — 全面 provider trait、FTS5 memory backend、3+ flows、visual editor MVP。
  2. M3 — marketplace、premium templates、健康 + 學習 flow。
  3. Post-M3 — NT$199-499/月 Pro tier 公開上線、Hahow 課程。

Sibling scaffold (do not confuse)

~/Documents/GitHub/hailmary/phantom-flow/ 是更早的 launchd heartbeat (每日 ping),保留不動。本 repo 是 M1 W2-4 substantial merge 的所在地。

License

Apache-2.0. © 2026 Mark Lai (markl-a). 兩個 subtree-merged repo 原始為 MIT (ai_automation_framework/LICENSE, data_analysis/LICENSE),與 Apache-2.0 相容。

About

Alpha · Tier 1 — Event-driven, cluster-aware workflow engine on phantom-mesh. Self-hosted n8n/Zapier alternative — Rust+Python, AI-native, local-first.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors