lfx-liam-bundle

A Langflow Extension implementing full GraphRAG aligned with the Microsoft GraphRAG default dataflow.

Backends: AstraDB, ArangoDB.

What you get

Stage	Capability
Index	Token chunking (default 1200 / overlap 100) → TextUnits
Index	Standard: LLM entity/relationship + Data Gleaning
Index	FastGraphRAG: NLP noun phrases + co-occurrence (cheaper/noisier)
Index	Hierarchical Leiden communities (+ Louvain fallback)
Index	Community reports (generate + summarize); optional Claims
Index	Embeddings + bidirectional provenance
Index	Backend vector ANN indexes (default on)
Query	Local Search (ANN entity entry + subgraph load)
Query	Global Search map-reduce
Query	DRIFT Search (community primer + local follow-ups)

Components

Side	Component
Build	GraphRAG Knowledge Base / Index Builder
Query	GraphRAG Retrieve (Local / Global / DRIFT)
Ops	GraphRAG Maintain
Provenance	GraphRAG Provenance

Install

pip install lfx-liam-bundle
# or into local Langflow docker:
./scripts/deploy-to-docker.sh

Python 3.10+. Search UI for GraphRAG / Liam.

ArangoDB requirements & troubleshooting

If you choose ArangoDB, read this before blaming the Flow.

Server requirements

Requirement	Why
ArangoDB ≥ 3.12.4 (3.12.6+ recommended)	Vector AQL + index APIs
Vector index startup flag	`3.12.4`: `--experimental-vector-index true`; newer may use `--vector-index`
Network reachability from Langflow to `arango_url`	Connection / timeout errors otherwise
User with rights to create collections, graphs, indexes	Schema + ANN setup

Enable vector indexes (examples):

arangod --experimental-vector-index true   # 3.12.4
# arangod --vector-index true              # newer builds

Local one-liner:

./devops/db-up.sh
./devops/test-integration.sh   # real DB tests (Arango required; Astra if env set)
./devops/db-down.sh

Enabling vector indexes permanently changes storage layout. Tiny corpora (<40 docs) use IVF{n},Flat automatically (HNSW can SIGSEGV on Arango 3.12.4).

What this bundle does on Arango

Creates document collections: {prefix}_chunks/_entities/_relationships/_communities/_reports/_covariates/_documents
Creates graph {prefix}_kg_graph with {prefix}_entity_edges
After indexing (ANN on): creates Faiss vector indexes on
- entities.description_embedding
- chunks.embedding
- reports.embedding
Default factory template: IVF100_HNSW10,Flat — IVF base + HNSW; IVF list count is auto-shrunk to fit document count (small corpora will not use literal IVF100)
Local/DRIFT entity/report entry uses AQL APPROX_NEAR_COSINE (or L2 / innerProduct)

Symptom → check → fix

Symptom (UI / log)	Likely cause	What to do
Vector index create fails / unknown type	Flag off or old version	Enable `--experimental-vector-index` / `--vector-index`; rebuild
Arango Exit 139 / SIGSEGV on index create	Tiny corpus + HNSW factory	Upgrade this bundle (auto Flat for <40 docs) or newer Arango
`APPROX_NEAR_*` AQL error	No vector index, or metric mismatch	Rebuild with ANN on; keep metric=`cosine` unless you know you need l2/IP
`向量ANN=失败(将回退精确余弦)`	Index create failed but fallback allowed	Fix server (above); or temporarily rely on fallback
ANN 直接报错、不回退	“ANN 失败回退精确余弦” turned off	Turn it back on, or fix Arango first
`集合不存在`	Wrong prefix / DB / user	Same `知识库前缀名` + database as build; toggle “不存在则创建”
Local 慢但 ANN 显示 ready	Subgraph path fell back to full load	Check `index_load` in retrieve meta (`subgraph` vs `full`)
维度不一致	Different Embedding model than build	Use the same Embedding; or rebuild with overwrite
认证失败 / 401	Wrong user/password	Fix Arango credentials on KB component
小库建索引怪错	`nLists` > doc count (should be rare)	Rebuild; bundle auto-shrinks nLists — upgrade bundle if old

Minimal Arango checklist

curl -s http://<host>:8529/_api/version works from Langflow host
Vector feature enabled (--vector-index)
KB component: URL / DB / user / password / prefix correct; 启用向量库 ANN 检索=开
Build with 覆盖重建 once after enabling vector feature
Retrieve meta should show vector_ranking=ann:arangodb and preferably index_load=subgraph

Differences vs Microsoft GraphRAG

Topic	Microsoft GraphRAG	This bundle
Chunking	Built-in token chunks	✅
Standard extract	LLM + gleaning	✅
FastGraphRAG index	NLP nouns + co-occurrence	✅ (lightweight regex NLP; no spaCy required)
Communities	Hierarchical Leiden	✅ Leiden (+ Louvain fallback)
Local / Global / DRIFT	Official engines	✅ All three (compact Chinese-oriented prompts)
Context budget	Token packing	✅ tiktoken
Vector ANN	Dedicated vector store	✅ Astra `$vector` / Arango Faiss IVF(+HNSW)
Subgraph load after ANN	Vector store + selective load	✅ `load_subgraph` after ANN seeds
Prompt-tuning CLI	`graphrag` CLI / prompt tune	❌ Intentionally omitted — tune via Langflow component params + prompts in code

Prompt / CLI note

Microsoft’s prompt-tuning CLI is a standalone repo workflow. This extension runs inside Langflow: response type, history, budgets, DRIFT depth, indexing mode, ANN toggles are component inputs. Full CLI parity is not a goal; if you need Microsoft’s exact long English prompt packs, fork and replace prompt strings under graphrag/*.

Docs

Usage (Chinese)
Architecture (Chinese)
Development (Chinese)
Upstream: Indexing · Local · Global · DRIFT · Methods / FastGraphRAG

License

MIT

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

lfx_liam_bundle-0.0.1.tar.gz (63.3 kB view details)

Uploaded Aug 9, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

lfx_liam_bundle-0.0.1-py3-none-any.whl (71.0 kB view details)

Uploaded Aug 9, 2026 Python 3

File details

Details for the file lfx_liam_bundle-0.0.1.tar.gz.

File metadata

Download URL: lfx_liam_bundle-0.0.1.tar.gz
Upload date: Aug 9, 2026
Size: 63.3 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for lfx_liam_bundle-0.0.1.tar.gz
Algorithm	Hash digest
SHA256	`c5250d5c336122d67db8f8c1863c145742727889ba8a496ea2bed7af36bd8c31`
MD5	`ee01e2b921aa7c0380d569edf856166e`
BLAKE2b-256	`154554a1974f8155ae11362e0599db48451037c411518c2778de11786c73d20c`

See more details on using hashes here.

Provenance

The following attestation bundles were made for lfx_liam_bundle-0.0.1.tar.gz:

Publisher: python-publish.yml on loadingvx/lfx-liam-bundle

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: lfx_liam_bundle-0.0.1.tar.gz
- Subject digest: c5250d5c336122d67db8f8c1863c145742727889ba8a496ea2bed7af36bd8c31
- Sigstore transparency entry: 2387949688
- Sigstore integration time: Aug 9, 2026
Source repository:
- Permalink: loadingvx/lfx-liam-bundle@7e7e16b90a031c825605fa478eaa2fc845534b7c
- Branch / Tag: refs/tags/v0.0.1
- Owner: https://github.com/loadingvx
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: python-publish.yml@7e7e16b90a031c825605fa478eaa2fc845534b7c
- Trigger Event: release

File details

Details for the file lfx_liam_bundle-0.0.1-py3-none-any.whl.

File metadata

Download URL: lfx_liam_bundle-0.0.1-py3-none-any.whl
Upload date: Aug 9, 2026
Size: 71.0 kB
Tags: Python 3
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for lfx_liam_bundle-0.0.1-py3-none-any.whl
Algorithm	Hash digest
SHA256	`6294b8e9c431dad88811e6371cd974e33ed0080b5379cbdc2ce0e924c4d6b6f9`
MD5	`0f2461742f5507e22750f29a1c0adb5b`
BLAKE2b-256	`81dce993864904274d2907c0430697231703e064ece8ca065f1c1e247ef95070`

See more details on using hashes here.

Provenance

The following attestation bundles were made for lfx_liam_bundle-0.0.1-py3-none-any.whl:

Publisher: python-publish.yml on loadingvx/lfx-liam-bundle

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: lfx_liam_bundle-0.0.1-py3-none-any.whl
- Subject digest: 6294b8e9c431dad88811e6371cd974e33ed0080b5379cbdc2ce0e924c4d6b6f9
- Sigstore transparency entry: 2387949692
- Sigstore integration time: Aug 9, 2026
Source repository:
- Permalink: loadingvx/lfx-liam-bundle@7e7e16b90a031c825605fa478eaa2fc845534b7c
- Branch / Tag: refs/tags/v0.0.1
- Owner: https://github.com/loadingvx
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: python-publish.yml@7e7e16b90a031c825605fa478eaa2fc845534b7c
- Trigger Event: release

Release history Release notifications | RSS feed

0.0.3

Aug 9, 2026 2 files

This release

0.0.1 This release

Aug 9, 2026 2 files

lfx-liam-bundle 0.0.1

lfx-liam-bundle

What you get

Components

Install

ArangoDB requirements & troubleshooting

Server requirements

What this bundle does on Arango

Symptom → check → fix

Minimal Arango checklist

Differences vs Microsoft GraphRAG

Prompt / CLI note

Docs

License

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License

Requires

Tags

Classifiers

Download files

Source Distribution

Built Distribution

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

Release history Release notifications | RSS feed