Add promotion launch kit

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,50 @@
+name: Bug report
+description: Report a reproducible problem with synaptic-memory.
+title: "[Bug]: "
+labels: ["bug"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for taking the time to report this. A small reproduction is the fastest path to a fix.
+  - type: textarea
+    id: summary
+    attributes:
+      label: Summary
+      description: What went wrong?
+    validations:
+      required: true
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: Reproduction
+      description: Paste the smallest code sample or commands that reproduce the issue.
+      render: bash
+    validations:
+      required: true
+  - type: textarea
+    id: expected
+    attributes:
+      label: Expected behavior
+    validations:
+      required: true
+  - type: textarea
+    id: environment
+    attributes:
+      label: Environment
+      description: Python version, synaptic-memory version, OS, backend, and installed extras.
+      value: |
+        - Python:
+        - synaptic-memory:
+        - OS:
+        - Backend:
+        - Extras:
+    validations:
+      required: true
+  - type: textarea
+    id: logs
+    attributes:
+      label: Logs or trace
+      render: text
+    validations:
+      required: false

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: true
+contact_links:
+  - name: Questions and adoption help
+    url: https://github.com/PlateerLab/synaptic-memory/discussions
+    about: Ask questions, discuss integration choices, or share use cases.
+  - name: Documentation
+    url: https://github.com/PlateerLab/synaptic-memory#readme
+    about: Start with the README, quickstart, install extras, and reports.

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,41 @@
+name: Feature request
+description: Suggest a capability, backend, integration, or documentation improvement.
+title: "[Feature]: "
+labels: ["enhancement"]
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem
+      description: What workflow or limitation are you trying to solve?
+    validations:
+      required: true
+  - type: textarea
+    id: proposal
+    attributes:
+      label: Proposed solution
+      description: What would you like synaptic-memory to do?
+    validations:
+      required: true
+  - type: dropdown
+    id: area
+    attributes:
+      label: Area
+      options:
+        - Ingest
+        - Search / ranking
+        - MCP / agent tools
+        - Memory operating layer
+        - Backend / infrastructure
+        - Documentation
+        - Evaluation / benchmarks
+        - Other
+    validations:
+      required: true
+  - type: textarea
+    id: context
+    attributes:
+      label: Additional context
+      description: Links, examples, corpus shape, scale, or constraints.
+    validations:
+      required: false

README.ko.md

@@ -24,6 +24,24 @@ synaptic-quickstart --db quickstart.db
 
 ---
 
+## 왜 그냥 RAG가 아닌가?
+
+일반 RAG는 독립적인 chunk에서 답을 찾는 경우가 많습니다. Synaptic은 먼저
+그래프를 만들기 때문에 에이전트가 검색하고, 관계를 따라가고, 정형 row를
+조회하고, 어떤 evidence가 도움이 됐는지 기억할 수 있습니다.
+
+| 일반 RAG | Synaptic Memory |
+|----------|-----------------|
+| chunk + vector search | 문서, chunk, row, edge |
+| 데이터 변경 시 재빌드가 흔함 | live DB용 CDC sync |
+| 기본은 single-shot retrieval | 멀티턴 탐색용 MCP 도구 |
+| feedback은 인덱스 밖에 있음 | 선택적 memory event, feedback, health signal |
+
+vector DB 대체재가 아닙니다. 기존 문서, SQL 데이터, embedding endpoint,
+agent runtime 주변에 붙는 graph/tool layer입니다.
+
+---
+
 ## 그래프 구축과 검색
 
 ```python

README.md

@@ -26,6 +26,24 @@ zero-dependency smoke test. Full source for the expanded example:
 
 ---
 
+## Why not just RAG?
+
+Plain RAG usually answers from independent chunks. Synaptic builds a graph first,
+so an agent can search, follow relations, inspect structured rows, and remember
+which evidence helped.
+
+| Plain RAG | Synaptic Memory |
+|-----------|-----------------|
+| chunks + vector search | documents, chunks, rows, and edges |
+| usually rebuilds on data change | CDC sync for live databases |
+| single-shot retrieval by default | MCP tools for multi-turn exploration |
+| feedback lives outside the index | optional memory events, feedback, and health signals |
+
+It is not a vector database replacement. It is the graph and tool layer around
+your existing documents, SQL data, embedding endpoint, and agent runtime.
+
+---
+
 ## Build and search
 
 ```python

docs/launch/promotion_playbook.md

@@ -0,0 +1,202 @@
+# Promotion Playbook
+
+This is the working launch plan for Synaptic Memory. Keep claims tied to the
+current README, reports, and reproducible commands.
+
+---
+
+## Positioning
+
+Short form:
+
+> Synaptic Memory is a Python knowledge graph + MCP tool server for agentic RAG:
+> deterministic indexing by default, infra-friendly backends, live DB sync, and
+> optional memory feedback.
+
+One-line GitHub/social variant:
+
+> Graph memory for RAG agents: documents + SQL rows + edges + MCP tools, with
+> LLM-free indexing by default.
+
+What to emphasize:
+
+- LLM-free deterministic default path.
+- Mixed structured and unstructured retrieval.
+- MCP tools for multi-turn agent search.
+- Live database sync.
+- Infra options: SQLite locally, PostgreSQL for shared service, Kuzu/Qdrant/MinIO
+  for scale-out composition.
+- Memory operating layer: feedback, provenance, scope-aware scores, health
+  signals.
+
+What not to claim:
+
+- Do not call it a vector database replacement.
+- Do not say it is 10TB-ready out of the box.
+- Do not imply the memory layer decides truth automatically.
+- Do not quote benchmark numbers without linking the report and reproduction
+  path.
+
+---
+
+## Launch Assets
+
+Must-have:
+
+- GitHub README with quickstart and "Why not just RAG?"
+- PyPI metadata links: repository, docs, changelog, issues, discussions, reports,
+  quickstart.
+- GitHub topics and description.
+- GitHub release notes.
+- Show HN draft: `docs/launch/show_hn_draft.md`.
+
+Nice-to-have:
+
+- 30-60 second terminal GIF:
+  `pip install`, `synaptic-quickstart --json`, `from_data()`, `search()`.
+- Architecture image showing documents/SQL -> graph -> EvidenceSearch -> MCP
+  agent tools.
+- Short Korean post for GeekNews / LinkedIn.
+
+---
+
+## GitHub Release Draft
+
+Title:
+
+> Synaptic Memory: graph memory layer for agentic RAG
+
+Body:
+
+~~~markdown
+Synaptic Memory is a Python knowledge graph + MCP tool server for agentic RAG.
+The default path is deterministic and LLM-free at indexing time, while optional
+layers add embeddings, reranking, OpenIE extraction, retrieval feedback, and
+memory health signals.
+
+Highlights:
+
+- `synaptic-quickstart` for a 5-minute local smoke test.
+- `SynapticGraph.from_data()`, `from_chunks()`, and `from_database()` presets:
+  `local`, `rag`, `agent`, `scale`.
+- Infrastructure paths for SQLite, PostgreSQL + pgvector, Kuzu, Qdrant, and
+  MinIO/S3-compatible blob storage.
+- MCP tools for deep search, comparison search, structured filters, aggregation,
+  FK joins, top-N queries, ingestion, and CDC sync.
+- Memory operating layer: retrieval events, feedback, provenance, scope-aware
+  reinforcement, and health reporting.
+
+Try it:
+
+```bash
+pip install "synaptic-memory[sqlite,korean,vector]"
+synaptic-quickstart --db quickstart.db
+```
+
+Reports:
+
+- RAG comparison: docs/REPORT-rag-vs-synaptic.md
+- Memory operating layer eval: docs/REPORT-memory-operating-layer-eval.md
+- Reproducible results log: docs/comparison/synaptic_results.md
+~~~
+
+---
+
+## Social Copy
+
+### X / Twitter
+
+```text
+I built Synaptic Memory: a Python knowledge graph + MCP server for agentic RAG.
+
+Default path:
+- no LLM calls at indexing time
+- local SQLite quickstart
+- optional PostgreSQL/Kuzu/Qdrant/MinIO infra
+- MCP tools for multi-turn search
+- feedback + memory health signals
+
+https://github.com/PlateerLab/synaptic-memory
+```
+
+### LinkedIn
+
+```text
+I have been building Synaptic Memory, a Python library for teams that need RAG
+over both documents and SQL data without turning every indexing job into an LLM
+job.
+
+It builds a deterministic knowledge graph by default, exposes MCP tools for
+agentic multi-turn search, supports live database sync, and can run locally on
+SQLite or connect to PostgreSQL, Kuzu, Qdrant, and MinIO/S3-style storage.
+
+The goal is not to replace vector databases. It is to provide the graph, tool,
+and memory layer around existing documents, SQL data, embedding endpoints, and
+LLM agents.
+
+GitHub: https://github.com/PlateerLab/synaptic-memory
+PyPI: https://pypi.org/project/synaptic-memory/
+```
+
+### Reddit
+
+Use only in communities where project posts are allowed. Lead with the problem,
+not the link.
+
+```text
+I built a Python knowledge graph + MCP server for RAG agents over mixed
+documents and SQL data.
+
+The main design choice: indexing is deterministic by default, so adding data
+does not require an LLM call. The agent can still use embeddings/rerankers and
+MCP tools at query time.
+
+Useful if you care about local/on-prem deployments, live DB sync, Korean FTS, or
+agent search over both structured and unstructured data.
+
+I would appreciate feedback on the tool surface and benchmark plan:
+https://github.com/PlateerLab/synaptic-memory
+```
+
+---
+
+## Channel Order
+
+1. GitHub release.
+2. PyPI release metadata check.
+3. HN Show HN.
+4. Korean mirror post: GeekNews / LinkedIn.
+5. Reddit posts only where allowed.
+6. Follow-up technical blog post:
+   "Why RAG agents need graph tools, not only vector search."
+
+---
+
+## Follow-Up Blog Outline
+
+Title:
+
+> Why RAG agents need graph tools, not only vector search
+
+Outline:
+
+1. Plain RAG is strong but chunk-local.
+2. Enterprise corpora mix documents, tables, foreign keys, policies, and updates.
+3. Index-time LLM extraction is useful but costly and hard for on-prem.
+4. Deterministic graph first: documents, chunks, rows, references, FK edges.
+5. Query-time agent tools: search, expand, filter, aggregate, join, top-N.
+6. Memory operating layer: feedback and provenance as metadata, not prompt bulk.
+7. Honest limits: not web-scale without an operating layer, not truth discovery.
+
+---
+
+## Measurement Backlog
+
+Good promotion gets stronger when each claim has a command behind it.
+
+- Clean-install quickstart timing.
+- Public Korean FTS benchmark.
+- Agent search benchmark with DeepSeek Flash profile.
+- Postgres CDC parity smoke.
+- CompositeBackend smoke with Kuzu + Qdrant + MinIO.
+- Large-corpus benchmark report linked from `docs/comparison/synaptic_results.md`.