isage 0.3.2.2

SAGE - Streaming-Augmented Generative Execution (meta package)

SAGE - Streaming-Augmented Generative Execution

A declarative, composable framework for building transparent LLM-powered systems through dataflow abstractions.

🚀 Quick Start

---

SAGE is a high-performance streaming framework for building AI-powered data processing pipelines. Transform complex LLM reasoning workflows into transparent, scalable, and maintainable systems through declarative dataflow abstractions.

2026 Focus Reset

SAGE is being refocused into a stream-first inference service system instead of a broad collection of loosely coupled apps.

• Keep the stream core: DataStream + declarative pipeline composition remain the product identity
• Keep the execution core: LocalEnvironment, JobManager, scheduling, service runtime
• Keep the serving integration plane: OpenAI-compatible gateway access, model lifecycle entry, and control-plane integration contracts
• Keep distributed execution optional: FluttyEnvironment remains the Flutty-first optional distributed runtime entry, instead of falling back to new Ray-based paths
• Keep the operating substrate: centralized ports, XDG user paths, model registry, logs, health/status surfaces
• De-emphasize apps: UI-first repos are no longer the product center; optional apps should sit outside the core or be retired

This direction also makes SAGE easier to position as a SAGE Zoo member: a reusable stream-oriented runtime + serving component that other systems can call through stable APIs instead of embedding internal implementation details.

Important boundary: isagellm remains an independent inference engine. SAGE integrates with it as an external engine/service capability; SAGE should not absorb isagellm internals into the main repository.

Preferred in-tree surface during consolidation:

• from sage.stream import DataStream
• from sage.runtime import LocalEnvironment, FluttyEnvironment, JobManager
• from sage.foundation import SagePorts, SageUserPaths
• from sage.serving import SageServeConfig, build_sagellm_gateway_command, probe_gateway

Key Features

• Stream-First: Dataflow is the primary abstraction, not an afterthought
• Production-Ready: Local-first execution, optional distributed processing, fault tolerance, comprehensive monitoring
• Developer Experience: Complex AI pipelines in just a few lines of code
• High Performance: Optimized streaming with intelligent memory management
• Observable: Built-in visibility into execution and performance
• Flexible: CPU-only or GPU nodes with intelligent resource scheduling

Quick Start

Transform rigid LLM applications into flexible, observable workflows. Traditional imperative approaches create brittle systems:

# Traditional approach - rigid and hard to modify
def traditional_rag(query):
    docs = retriever.retrieve(query)
    if len(docs) < 3:
        docs = fallback_retriever.retrieve(query)
    prompt = build_prompt(query, docs)
    response = llm.generate(prompt)
    return response

SAGE transforms this into a declarative, composable workflow:

from sage.runtime import LocalEnvironment
from sage.libs.foundation.io.source import FileSource
from sage.middleware.operators.llm import SageLLMGenerator  # ✅ Recommended
from sage.libs.foundation.io.sink import TerminalSink

# Create execution environment
env = LocalEnvironment("rag_pipeline")

# Build declarative pipeline with sageLLM (recommended)
(
    env.from_source(FileSource, {"data_path": "questions.txt"})
    .map(SageLLMGenerator, {
        "model_path": "Qwen/Qwen2.5-7B-Instruct",
        "backend_type": "auto",  # auto/cuda/ascend/mock
    })
    .sink(TerminalSink)
)

# Execute pipeline
env.submit()

💡 LLM Engine: SAGE uses sageLLM as the default inference engine. For OpenAI-compatible APIs, use OpenAIGenerator. See CHANGELOG for legacy migration notes.

Current API quick reference

• sage.libs.foundation.io.source: FileSource, TextFileSource, CSVFileSource, JSONFileSource
• sage.libs.foundation.io.sink: TerminalSink, FileSink
• sage.middleware.operators.rag: RAGDocument, RAGQuery, RAGResponse
• sage.middleware.operators.llm: SageLLMGenerator

Try it yourself

git clone https://github.com/intellistream/SAGE.git && cd SAGE
git checkout main-dev
./quickstart.sh --dev --yes

# Tutorials are now in a separate repository
git clone https://github.com/intellistream/sage-tutorials.git
python sage-tutorials/L1-common/hello_world.py

For CPU-only verification

# Verify the in-tree core surface
sage verify

# Run a real one-shot chat via sagellm (gateway or direct CLI)
sage chat --ask "Hello, SAGE!"

# Explore tutorials separately if needed
git clone https://github.com/intellistream/sage-tutorials.git
python sage-tutorials/L1-common/hello_world.py

Architecture

Current baseline is a 4-tier workspace architecture (L1-L4):

L4: application repos (optional)      # App / UI / benchmark
L3: sage.cli                          # CLI entrypoint        (in-tree)
L2: sage.runtime + sage.stream        # Runtime / scheduler   (in-tree)
L1: sage.foundation                   # Foundation            (in-tree)

Notes:

• Historical split packages may still exist as transitional published compatibility channels.
• The main repository now owns the preferred product surface directly: sage.foundation + sage.stream + sage.runtime + sage.serving + sage.cli.
• Historical split foundation/runtime/CLI repos are therefore no longer the desired long-term product boundary, even when some transitional imports still exist outside the main install contract.

Target product convergence is narrower than the historical workspace shape:

SAGE Inference Service System
L3 Interface   : CLI + OpenAI-compatible service entry + external integration surface
L2 Runtime     : LocalEnvironment + DataStream + JobManager + scheduler + execution services
Optional Dist. : FluttyEnvironment (Flutty-backed distributed execution)
L1 Foundation  : config + ports + user paths + model registry + logging
Optional       : RAG / memory / tool-use / benchmark adapters

In other words, SAGE is moving toward a smaller, sharper center: stream + runtime + serving + operations, with distributed execution available as an optional scale-out mode.

Repo-retirement gate: do not retire historical split repos solely based on packaging cleanup. The main repo has now removed direct historical runtime split-package dependency pins and owns its local runtime path in-tree, but ecosystem compatibility imports, external repos, and remaining transitional release channels still need deliberate follow-up before those repos can be fully retired.

See SAGE Ecosystem for all independent sub-repositories with CI status, PyPI packages, and categorized listings.

📖 Architecture Guide - Canonical ownership boundaries and dependency rules for the meta repo

📌 Layer Ownership Matrix v1 (Wave A) - Canonical L1-L4 workspace ownership, independent sub-repo coordination boundary (including sagellm capabilities), forbidden directions, and boundary refactor review checklist

Installation

Quickstart (Recommended)

git clone https://github.com/intellistream/SAGE.git && cd SAGE
./quickstart.sh --dev --yes         # 开发模式：尽量本地 editable
# 或
./quickstart.sh --standard --yes    # 标准模式：子包依赖默认从 PyPI 安装

⚡ Auto-Acceleration: Network optimization is now enabled by default:

• 🌐 Auto-detects network location (China mainland → mirror sources)
• 🚀 Parallel downloads (8 threads) + pre-compiled packages
• ⏱️ 3-5x faster installation: 12-18 min (vs 35-45 min)
• 🔧 Disable: ./quickstart.sh --no-mirror --dev --yes

Install Mode Semantics

• standard：本地安装仓库根目录下的 isage meta 包，子仓依赖按根 pyproject.toml 版本从 PyPI 解析。
• dev：先完成 standard 安装，再尽量将同级工作区中的本地 SAGE 子仓库切换为 editable (-e)。

PyPI Install

pip install isage              # Core framework
pip install isage[dev]         # Development tools (includes isage-dev-tools, pre-commit, pytest, etc.)

What's included in pip install isage

isage now ships the main product surface directly from this repository: sage.foundation + sage.stream + sage.runtime + sage.serving + sage.cli. The default local execution path no longer requires a separate historical runtime split package as a direct dependency. isagellm remains the external inference engine.

On Python 3.13+, the root package currently skips automatic isagellm installation because the required isagellm-protocol distribution is not yet published for that interpreter. Core SAGE stream/runtime development still installs normally; engine integration can be enabled later on a supported interpreter once upstream wheels are available.

Core + engine integration only 🧩

For a standard pip install isage, the product center is intentionally narrow:

• Foundation → isage: in-tree config, ports, paths, contracts
• Stream + Runtime → isage: main public API and local runtime owned in-tree
• Distributed scale-out → flutty: optional backend used through FluttyEnvironment
• CLI → isage: in-tree sage command surface
• Inference Engine → isagellm: external engine; auto-installed on supported Python versions

Compatibility note: transitional imports such as sage.common, sage.platform, sage.middleware, or sage.kernel may still appear in older repos or environments, but they are not part of the root package's direct dependency contract anymore.

Edge aggregation now also lives in-tree as sage.edge; install isage[serving-edge] or isage[full] to use the sage-edge shell.

Optional adapter packages 🦁

These packages are no longer part of the default isage install. They remain independent optional adapters and can be installed explicitly or via isage[full] when needed.

Policy note: optional adapters should justify their independence with real owned functionality. Thin wrapper repos should be folded back into the main SAGE repository instead of expanding the default dependency surface.

Current example: sage.edge has already been folded back into the main repo. The former sage-edge split repo should be treated as retired rather than as an independent Zoo package.

pip install 'isage[serving-edge]'         # in-tree edge shell（sage.edge / sage-edge）
pip install 'isage[capability-adapters]'  # intent / rag / neuromem adapters
pip install 'isage[capability-tooluse]'   # SIAS tool-use adapter
pip install 'isage[full]'                 # all optional adapters + data package

sage-edge --port 8899            # 挂载外部 sagellm gateway 的 edge shell
pip install isage-rag              # RAG 管道（文档加载 / 分块 / 检索 / 重排）
pip install isage-neuromem         # 记忆 / 检索持久化
pip install isage-libs-intent      # 意图识别（关键词 + LLM）
pip install isage-eval             # 评估框架（指标 / LLM 评判）
pip install isage-finetune         # LLM 微调 / Agent training（LoRA / SFT / RL / Reward Model）
pip install isage-agentic-tooluse  # Agent 工具选择（Hybrid/DFS/Gorilla）
pip install 'isage-tools[mcp]'     # 独立工具仓库；当前已注册到 sage-mcp
pip install 'isage-mcp[all]'       # 聚合 MCP Server（当前默认聚合 isage-tools）

Example: install full core SAGE stack

pip install isage

See Dependency Management in DEVELOPER.md for detailed guidance.

Verification & Troubleshooting

sage doctor                    # Check installation
sage verify                    # Verify in-tree core surface
sage chat --help               # Inspect chat entrypoints
sage index ingest --help       # Inspect lightweight index entrypoints
./quickstart.sh --doctor       # Diagnose issues

CLI Command Reference

The current main-repo sage command surface is intentionally small and grouped around the core product boundary:

Command	Purpose
sage version	Print installed SAGE version
sage status	Show local config/data/state paths and gateway summary
sage doctor	Run lightweight environment diagnostics
sage verify	Smoke-check the in-tree core surface
sage runtime nodes	List runtime-visible nodes
sage serve gateway --json	Print the external sagellm gateway launch contract
sage serve gateway --probe --json	Probe the configured gateway health endpoint
sage chat	Start chat via sagellm gateway, direct CLI, or configured OpenAI-compatible backend
sage chat --ask "..."	Run one-shot chat
sage index ingest --source ./docs --index local-docs	Record lightweight local index metadata

sage verify
sage runtime nodes
sage serve gateway --json
sage chat --ask "Hello, SAGE!"
sage index ingest --source ./docs --index local-docs

📖 Detailed guides: Installation Guide | Troubleshooting | Validation | Optimization Tips

⚠️ Known Issues: If you encounter transformers version conflicts when installing multiple SAGE packages, prefer checking DEVELOPER.md and the package-specific READMEs first.

Environment Configuration

cp .env.template .env    # Copy template
# Edit .env and add your API keys (OPENAI_API_KEY, HF_TOKEN, etc.)

📖 API key setup: See .env.template for all available options

📚 Tutorials

Complete tutorials covering the current workspace tiers of SAGE (L1-L4) plus historical capability topics:

# Clone tutorials repository
git clone https://github.com/intellistream/sage-tutorials.git
cd sage-tutorials

# Start learning (30 seconds)
python L1-common/hello_world.py

# Follow the quick start guide
cat QUICK_START.md

Tutorial Structure:

• sage-tutorials/L1-common/ - Foundation layer (config, logging, unified client)
• sage-tutorials/L2-platform/ - Platform services (scheduler, storage)
• sage-tutorials/L3-kernel/ - Execution engine (batch, stream, operators)
• sage-tutorials/L3-libs/ - RAG, Agents, Algorithms
• sage-tutorials/L4-middleware/ - Domain operators (vector DB, time-series)
• sage-tutorials/L5-apps/ - Applications and integration demos

See sage-tutorials/README.md for complete learning paths.

Documentation & Resources

• Documentation: https://intellistream.github.io/sage-docs/
• Examples & Applications: intellistream/sage-examples
  • RAG examples and production applications
  • Will be published as isage-examples on PyPI
• Tutorials: intellistream/sage-tutorials
  • Layered tutorials from L1 to L5, quick-start learning paths
• Architecture: sage-docs architecture guide

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

git checkout -b feature/my-feature
./quickstart.sh --dev --yes
# Make changes, add tests
sage-dev quality && sage-dev test
git commit -m "feat(kernel): add new feature"
git push -u origin feature/my-feature

Resources: Quick Reference | GitHub Issues | Discussions

Team Information

🔒 Team assignments and sensitive information are maintained in a private repository to protect member privacy.

• Public: Project-level information is available in this repository
• Private: Team member assignments, funding details, and contact information are accessible to authorized members only
• Access: Contact project management for access to the private repository

Developer Tools

make help           # View all commands
sage-dev quality    # Format & lint
sage-dev test       # Run tests
make docs           # Build documentation

📖 Complete reference: DEVELOPER.md

SAGE Ecosystem

📦 sage-docs package guide — 独立能力包与安装索引

🧠 SAGE — Streaming AI Framework

SAGE is now centered on an in-tree core product surface rather than on a broad split-repo zoo.

Core Product Surface (owned in-tree)

• sage.foundation — config, ports, paths, logging, and shared contracts
• sage.stream — DataStream, transformations, operators, and flow composition
• sage.runtime — environments, scheduling, job management, and execution lifecycle
• sage.serving / sage.edge — serving integration boundary and edge aggregation shell
• sage.cli — main sage command surface

Legacy split repos and duplicate stream surfaces should be treated as consolidation/retirement targets rather than as Zoo members.

Independent sub-repositories that remain justified are organized by category:

Application & UI

• sage-examples — Tutorials and application examples

Algorithms & Libraries

• sage-agentic — ReAct, PlanExecute agents and agentic workflows
• sage-rag — Retrieval-augmented generation components
• sageVDB — High-performance vector database (FAISS-compatible API)
• sageRefiner — Query and response refinement algorithms
• sage-amms — Approximate matrix multiplication service
• flutty — Optional distributed execution backend for SAGE stream runtime

Data & Benchmarks

• sageData — Unified dataset management for SAGE subsystems
• sage-benchmark — Comprehensive evaluation framework for RAG, agents, memory, and control plane
• sage-eval — Evaluation metrics, profilers, and LLM judges

Model Optimization & Safety

• sage-finetune — Model fine-tuning, agent training, and adaptation
• sage-privacy — Differential privacy and PII handling
• sage-safety — Safety filters and guardrails

Developer Tooling

• sage-dev-tools — Development CLI and quality tooling
• Historical split repos remain retirement targets and are intentionally omitted from the recommended active ecosystem list.

---

⚡ sageLLM — LLM Inference Engine

sageLLM is a modular, high-performance LLM inference engine. All repositories are 🔒 private and published to PyPI.

Core Engine

• sagellm-core — Core inference engine
• sagellm-backend — Backend drivers (CUDA, Ascend)
• sagellm-protocol — Wire protocol and serialization

Gateway & Control

• sagellm-gateway — OpenAI-compatible API gateway
• sagellm-control-plane — Scheduling and resource management

Optimization

• sagellm-kv-cache — KV cache management
• sagellm-comm — Communication layer
• sagellm-compression — Compression algorithms

Tooling & Benchmarks

• sagellm-benchmark — Performance benchmarks
• sagellm-dev-tools — Development tooling

Community

💬 Join SAGE Community - WeChat, QQ, Slack, GitHub Discussions

License

SAGE is licensed under the MIT License.