Skip to content

STACK-MAP.md

Latest commit

 

History

History
68 lines (59 loc) · 5.01 KB

STACK-MAP.md

File metadata and controls

68 lines (59 loc) · 5.01 KB
Raw

STACK-MAP — ellmos-stack composition blueprint

Derived overview of what this stack is made of and which function each part serves. Canonical sources: stack.v2.json (machine-readable manifest, schema ellmos.stack.v2), docker-compose.yml, OPERATIONS.md, RELEASE_GATE.md. Update this map when the manifest or the compose file changes.

Role in the stack family

ellmos-stack is the public, self-hosted base layer ("deployment chassis") of the ellmos stack family: a local-first AI research kit. It is designed to be nested — a larger composition can declare this stack as a component instead of copying it. Every published stack in the family is listed in the catalog ellmos-ai/stacks.

Map

                         USER / AGENT (localhost only by default)
                                        │
 ┌─ ellmos-stack ── Docker Compose, pinned images, Linux ────────────────────────────┐
 │                                                                                   │
 │  ┌─ Inference ─────────────┐   ┌─ Automation ────────────┐                        │
 │  │ Ollama (docker)         │   │ n8n (docker)            │                        │
 │  │ role: local-inference   │   │ role: automation        │                        │
 │  └─────────────────────────┘   └─────────────────────────┘                        │
 │                                                                                   │
 │  ┌─ Knowledge / RAG ───────────────────┐   ┌─ Specialized state ──────────┐       │
 │  │ KnowledgeDigest (pinned-git)        │   │ USMC       memory.curated    │       │
 │  │ role: knowledge.search.default      │   │ GARDENER   memory.organic    │       │
 │  │ ingest → chunk → FTS5 → summarize   │   │ task-master tasks.default     │       │
 │  └─────────────────────────────────────┘   └──────────────────────────────┘       │
 │                                                                                   │
 │  ┌─ Included services (services/) ───────────────────────────────────────────┐    │
 │  │ research_pipeline  role: research   — paper search & processing           │    │
 │  │ auto_ingest        — feeds documents into KnowledgeDigest                 │    │
 │  │ process_summaries  — LLM summarization pass                               │    │
 │  │ telegram_gateway   — optional chat entry point                            │    │
 │  │ env_config         — environment/config plumbing                          │    │
 │  └───────────────────────────────────────────────────────────────────────────┘    │
 │                                                                                   │
 │  External literature APIs: pubmed · arxiv  (role: paper-search, https)            │
 └───────────────────────────────────────────────────────────────────────────────────┘
   Policies: local_first · network declared · max data sensitivity "user-local"
   Release gate: RELEASE_GATE.md + .github/workflows/release-gate.yml (pinned images)

Component roles (from stack.v2.json)

Component Kind Role Source
USMC module component memory.curated pinned-git
GARDENER module component memory.organic pinned-git
task-master module component tasks.default pinned-git
KnowledgeDigest module component knowledge.search.default (RAG) pinned-git
Ollama model-runtime local-inference docker
n8n workflow-engine automation docker
research-pipeline included-service research this repository
pubmed / arxiv literature-api paper-search https

Deliberately out of scope

  • Not a generic n8n starter kit, not an Open-WebUI-only setup, not a hosted agent platform (see README positioning).
  • No remote access without TLS/auth — services bind to localhost by default.
  • Multi-tenancy, SSO, billing and managed hosted operations are out of scope. This stack targets a single operator on their own Linux server.