convez.ai
MCP ServerAutomaçãoAvançado
DaveSteadman

KoreStack

DaveSteadman/KoreStack

Workspace de IA modular local com agentes, dados e ferramentas de código

Ver no GitHub
InicialPluginScore · de 0 a 100

Um espaço de trabalho de IA local-first construído como um sistema modular de serviços Python cooperativos. Integra um runtime de agente, armazenamento de conversa, camada de dados, editores de documento e código, hub de comunicações externas, redes visuais de agentes, agendamento e subsistemas de teste. Funciona como um único sistema supervisionado, nunca lançando subsistemas isoladamente.

Instalação

$npx convez-ai DaveSteadman/KoreStack

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
25%
Último commit
há 2 dias
Versão
v0.12
Issues abertas
0
Contribuidores
1

Casos de uso

  • Construir agentes IA que interagem com dados locais e repositórios
  • Automatizar fluxos de trabalho com agendamento e integração de comunicações
  • Indexar e buscar documentos, código e dados através de uma interface unificada
  • Testar e validar comportamentos de agentes em tempo real

Veredito

KoreStack é uma plataforma ambiciosa e bem-arquitetada para quem precisa de um sistema de IA local completo com múltiplos subsistemas integrados. Requer domínio técnico significativo (Python 3.11+, configuração de LLM, gestão de microserviços) e é mais adequado para engenheiros construindo soluções end-to-end do que para adoção rápida.

README

A tradução para o português deste README ainda está na fila.

KoreStack

KoreStack is a local-first AI workspace built from cooperating Python services: an agent runtime, a conversation store, a data layer, document and code editors, an external communications hub, visual agent networks, scheduling, and two test subsystems. It is started, supervised, and used as one system; you never launch a subsystem on its own.

KoreStack animated screenshots

Quick start

Requirements: Python 3.11+, a writable data root, and an Ollama host (or other configured LLM endpoint).

python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python .\main.py

Then open the landing page at http://127.0.0.1:29600/. It lists every service with its health, and links to each subsystem UI.

Before the first run review config/korestack_config.json (ports, paths.dataroot) and config/koreagent_config.json (LLM host and models). See config/README.md.

Command line

This is the only place the suite's command line is documented. The root main.py runs KoreStack/main.py.

CommandEffect
python .\main.pyStart every enabled service plus the landing page (same as start)
python .\main.py statusReport the health of the running services
python .\main.py --dry-runPrint the resolved start plan without launching anything
python .\main.py --services a,bStart only the named services (all by default)
python .\main.py --host H --ui-port POverride the bind host and landing-page port
python .\main.py --open-browserOpen the landing page once started

Valid service names: korechat, koreagent, koredatagateway, koredocs, korecode, korecomms, koreliveweb, korecron, koretest2, koreunittest, koreagentnetwork.

The subsystems

SubsystemPortWhat it doesHow it contributes
KoreStack29600Supervisor and landing pageStarts, probes, and links every other service; shows Ollama state
KoreAgent29601Agent runtime: sessions, tool loop, skills, slash commands, System OneThe brain; every other service that needs an LLM calls its API
KoreChat29602Canonical conversation, message, and event storeShared thread record for the agent and the comms layer
KoreData29603 + childrenGateway over feed, library, RAG, reference, graph, and scrape servicesOne search and retrieval surface for agents and browsers
KoreDocs29610Documents, spreadsheets, and diagrams on the user filesystemAgent-editable user content, via UI and MCP
KoreCode29611Browser code workspace with indexing and AI edit flowsCoding surface scoped to the repository
KoreComms29609Discord, Gmail, SFTP, and manual message bridgesIsolates external channels from the agent
KoreLiveWeb29613Web search, fetch, navigation, and Wikipedia MCP toolsLive web evidence, kept apart from local data
KoreCron29615Scheduled triggersTime-based starter for tests and named networks
KoreTest229616System tests: prompts and commands run against the live interfacesPer-build regression of the whole stack
KoreUnitTest29618Python unit tests, one file per testPer-build regression of internal code
KoreAgentNetwork29617Visual, executable networks of Python, LLM, and Decision blocksComposable automation that calls other services

Shared support:

FolderRole
KoreCommon/Path, config, logging, and service helpers imported by every service
KoreUI/Service-specific templates and static assets; UIElements is the shared shell and top bar
config/Checked-in suite configuration

Architecture

                     KoreStack (29600) supervisor + landing page
                                   | starts / probes /status
   +-----------+-----------+-------+-------+-----------+------------+
 KoreChat   KoreAgent    KoreData     KoreDocs/Code   KoreComms   KoreLiveWeb
 threads    LLM + tools  gateway      user content    channels    web tools
     ^         ^   ^         ^
     |         |   +---------+---- MCP tools, skills
     |         |
 KoreCron -> KoreAgentNetwork runs  KoreAgentNetwork -> /api/work-packet
 KoreCron -> Test/UnitTest sessions KoreTest2 judge  -> /api/work-packet
                                    KoreUnitTest     -> runs repo test files
  • Control plane: KoreStack resolves config, launches services, and polls each /status.
  • Agent runtime: KoreAgent owns orchestration and tools. KoreChat owns durable conversation state. KoreComms owns external channels.
  • Data services: KoreData and KoreDocs stay domain services that the agent reaches through MCP and HTTP, never internal libraries.
  • LLM models: a chat model plus an Ollama System One decision model (typed questions scored with probabilities). KoreAgent's /api/work-packet exposes both to the rest of the suite.
  • Automation and testing: KoreCron triggers time-based work; KoreAgentNetwork builds visual pipelines; KoreTest2 and KoreUnitTest verify each new build.

Shared service contract

  • / or /ui: browser shell
  • /api/...: JSON API
  • /status: health probe used by KoreStack
  • /mcp: MCP tools, where a service exposes them

Browser pages share the KoreUI/UIElements shell and top bar, which links all suite UIs.

Data layout

LocationPurpose
<dataroot>/datacontrol/Service-owned state: databases, schedules, logs, test cases and results
<dataroot>/datauser/User content: documents, sheets, diagrams, files that networks and agents read and write

<dataroot> is paths.dataroot in config/korestack_config.json, overridden by KORE_SUITE_DATAROOT. The repo Data/ folder is a reference layout only.

Troubleshooting

ProblemWhat to check
Start fails immediatelyActivate the venv and rerun pip install -r requirements.txt
A service is red on the landing pagepython .\main.py --dry-run for ports and enablement, then that service's log in <dataroot>/datacontrol/logs
Missing folders or databasespaths.dataroot must point to a writable location
Model calls failCheck config/koreagent_config.json and that the Ollama host and models exist
Unstyled pagesKoreUI/UIElements/ must be present
One service blocks the restStart fewer with --services ...

Documentation rule

Primary documentation lives in the root and subsystem READMEs. ChangeLog.md records dated changes.

Project-wide intent and architecture are defined in Requirements.md and Design.md. Architectural changes that alter a documented trust boundary, service contract, persistent data format, or model policy require an ADR under docs/adr/.

Ver o original no GitHub

Bom para

Relacionados