convez.ai
MCP Server
venkathub

atlas-compliance-copilot

venkathub/atlas-compliance-copilot

Ver no GitHub
InicialPluginScore · de 0 a 100

Permission-aware, cost-routed, evaluated enterprise AI copilot for a financial/compliance domain (Spring AI + LangGraph + MCP).

Instalação

$npx convez-ai venkathub/atlas-compliance-copilot

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
42%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

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

Atlas

Production-grade, permission-aware enterprise AI operations copilot for a financial/compliance domain.

An employee asks a question → Atlas retrieves only documents they are cleared to see → answers with citations → and, on request, executes governed actions behind a human-in-the-loop checkpoint. Every model call is cost-routed, evaluated, and traced.

Status: P0 (foundations) in progress. See docs/ROADMAP.md for the phase plan.


Architecture (6 subsystems + cross-cutting evals/observability)

ModuleStackResponsibilityPhase
ui/ReactChat + admin; streamed answers, inline citations, agent traceP5
gateway/Spring BootAuth, cost-aware model router, semantic cache, rate limitingP3
rag-engine/Spring AIPermission-aware retrieval: chunk/embed, hybrid search, citationsP1
(pgvector)Postgres + pgvectorEmbeddings + role-tagged metadata (RBAC), agent checkpointsP1/P4
agents/Python / LangGraphPlanner–executor agents, memory, human-in-the-loopP4
mcp-tools/Spring BootGoverned enterprise actions exposed over MCPP4
evals/Python / RAGAS/DeepEvalEval harness (CI-gated) + Langfuse/Grafana observabilityP2
infra/Docker / CICompose stack, multi-arch images, GitHub ActionsP0

The LLM never runs locally — it lives on a remote Ollama GPU endpoint set via OLLAMA_BASE_URL (see docs/RUNBOOK.md). All model config is env-swappable.

Quickstart (P0 — fills in as increments land)

cp .env.example .env        # then edit values (never commit .env)
# docker compose up         # (Increment 2) Postgres+pgvector + Redis
# make health               # (Increment 2) assert services healthy
# mvn -pl rag-engine test   # (Increment 3) unit tests; live smoke test is @Tag("live")

Documentation

License

Apache-2.0 © 2026 Venkatesh. Note: the FinanceBench dataset used from P1 stays under its own CC-BY-NC-4.0 license (see docs/DECISIONS.md ADR-0004) — that covers data, not this code.

Ver o original no GitHub

Bom para

Relacionados