HKUDS/LightRAG v1.5.7 on GitHub

Important Notes

New Feature: Add end-user-facing entry point to WebUI

v1.5.7 adds a second, end-user-facing entry — /workspace — together with the branding and prompt controls a deployment needs before it can be handed to real users.

A query-only entry for end users: /workspace
Customizable welcome, login and empty-state pages
A global USER_PROMPT_PREFIX

/workspace UI shows the chat surface and nothing else — no document management, no knowledge graph, no API-docs links — and is built for mobile use. Point UI_TEMPLATES_DIR at a small directory of Markdown files plus a manifest.json, and LightRAG serves your welcome page, login-page blurb, query empty state, brand logo and copyright line — in several languages, without rebuilding the WebUI. A locale that declares both a login blurb and an agreement document also switches on a login consent checkbox. A ready-to-copy bundle ships in docs/ui_templates_example/.

USER_PROMPT_PREFIX (or USER_PROMPT_PREFIX_FILE for long, multi-paragraph text) prepends deployment-wide instructions to every request's user_prompt. It gives the operator one place to set the LLM's role, tone and output format — Mermaid diagrams, LaTeX delimiters, citation style — for both UI entries and the API alike. Requests can never read or replace it; only the API field disable_user_prompt_prefix opts out.

For more information, please refer to:

docs/LightRAG-API-Server.md — § The /workspace Query Entry
docs/UserDefinedUI.md — the complete UI customization guide, plus docs/ui_templates_example/
README.md — USER_PROMPT_PREFIX / USER_PROMPT_PREFIX_FILE

What's New

Add an offline tool to migrate a graph from Apache AGE to PostgreSQL tables by @ysys143 in #3620
feat: support graph-first ingestion with deferred vector indexing by @letsgetai in #3473
feat: add explicit custom chunking selector by @er-s-an in #3665
feat(api): add ENABLE_API_DOCS switch for API documentation exposure by @danielaskdd in #3672
perf(nano): defer vector deletes to flush time by @0xPabloxx in #3680
perf(faiss): defer vector deletes to flush time by @yzxcj797 in #3693
feat(Bedrock): Allow Bedrock default credential providers by @a-bruhn in #3726
feat(ollama): report token usage through the TokenTracker contract by @YashvantHange in #3747
feat(webui): standalone /workspace query entry for end users (dual-entry WebUI) by @danielaskdd in #3729
feat(webui): add a login-page consent gate driven by the UI bundle by @danielaskdd in #3755
feat(webui): use WEBUI_TITLE as the browser tab title by @danielaskdd in #3763
Add user-defined UI customization bundle support (UI_TEMPLATES_DIR) by @danielaskdd in #3765
Use custom logo and deployment title on login page by @danielaskdd in #3768
feat(webui): customer-defined copyright line at the foot of the pre-login pages by @danielaskdd in #3773
feat(query): Add global user prompt prefix feature with cache partitioning by @danielaskdd in #3781
refactor: remove the deprecated GET /documents listing and get_docs_by_status by @danielaskdd in #3788

What's Changed

fix(kg): return copies from NetworkXStorage get_node and get_edge by @santhreal in #3588
fix(operate): compute most common entity type and normalize keywords on rebuild by @santhreal in #3589
fix(kg): return boolean values from get_all_update_flags_status by @santhreal in #3590
fix(utils): split GRAPH_FIELD_SEP in merge_source_ids by @santhreal in #3592
fix(utils): correct docstring example and harden compute_incremental_chunk_ids by @santhreal in #3591
fix(parser): bound native markdown image downloads per document (GHSA-25c3-j78v-83qx) by @danielaskdd in #3605
fix(kg): hydrate track docs via shared row helper and catch TypeError by @santhreal in #3595
fix(kg): make get_knowledge_graph truncation deterministic across graph backends by @santhreal in #3593
fix(security): bound DOCX decompression and contain sidecar image paths by @danielaskdd in #3614
fix(rebuild_vdb): drop lossy agtype quote-strip that corrupts quote-bearing entity_ids by @ysys143 in #3587
fix(extraction): surface token-limit truncation and fail on empty truncated responses by @pengpengyi92 in #3607
build(deps): bump the github-actions group with 2 updates by @dependabot[bot] in #3616
fix(parser): enforce OOXML member and DOCX image budgets by @danielaskdd in #3619
fix(graph): validate attributes on the manual entity/relation APIs (GHSA-c922-pw4m-4wcv) by @danielaskdd in #3623
fix(kg): enforce the graph-attribute contract in the backends that need it by @danielaskdd in #3624
fix(api): reject unstorable characters in a document source by @danielaskdd in #3626
fix(graph): preserve self-loops on entity rename by @pengpengyi92 in #3622
fix(graph): reject self-loop relations at the manual write paths by @danielaskdd in #3625
perf(kg): drop the seed-degree round trip from PGTable traversal by @ysys143 in #3618
fix(milvus): page vector/id lookups to respect the gRPC response ceiling #3584 by @El-Tatane in #3629
fix: forward embedding_dim to Ollama's embed API by @Vishwamitra in #3604
fix: disable thinking for Ollama extract/keyword roles by default by @Vishwamitra in #3599
fix(base): tolerate undeclared fields when rehydrating DocProcessingStatus by @qianheng-aws in #3633
fix(llm): close OpenAI stream/client when GeneratorExit interrupts a mid-COT close by @AmirF194 in #3635
fix(api): fail fast for custom Ollama embedding dimensions by @pengpengyi92 in #3636
build(deps-dev): bump the build-tools group in /lightrag_webui with 5 updates by @dependabot[bot] in #3641
build(deps): bump the react group across 1 directory with 3 updates by @dependabot[bot] in #3639
build(deps): bump lucide-react from 1.26.0 to 1.29.0 in /lightrag_webui in the ui-components group by @dependabot[bot] in #3640
build(deps): bump the content-rendering group across 1 directory with 2 updates by @dependabot[bot] in #3642
build(deps): bump the frontend-minor-patch group across 1 directory with 4 updates by @dependabot[bot] in #3643
fix(kg): refuse PGGraphStorage on Apache AGE >= 1.8.0, which crashes the backend by @ysys143 in #3621
Refactor/parser cancellation exception rename by @MasaSN in #3637
refactor(neo4j): clarify batch edge read contract by @danielaskdd in #3648
Add an offline tool to migrate a graph from Apache AGE to PostgreSQL tables by @ysys143 in #3620
fix(api): preserve source files when deleting duplicate markers by @pengpengyi92 in #3632
docs: fix broken example links in programming guide by @Shizoqua in #3649
docs: fix outdated defaults and env var names by @Shizoqua in #3653
fix(parser): bound external result downloads for MinerU and Docling by @Shizoqua in #3638
fix(webui): harden graph type color resolution against reserved property names by @danielaskdd in #3655
docs(env.example): add the missing EXTRACT role variables by @torvalstrom in #3651
fix(lmdeploy): preserve do_sample setting by @Shizoqua in #3654
fix(parser): prevent quadratic Markdown image parsing by @danielaskdd in #3656
feat: support graph-first ingestion with deferred vector indexing by @letsgetai in #3473
fix(api): generalize EMBEDDING_DIM startup guard to all embedding bindings by @ClaireXi99 in #3646
docs: document OrcaRouter as an OpenAI-compatible LLM binding by @Marc-oss-hub in #3650
fix: warn when selectors bypass custom chunking by @er-s-an in #3662
fix(chunker): reject negative token overlap by @Shizoqua in #3663
fix(rerank): reject negative chunk overlap by @Shizoqua in #3667
fix(lollms): honour configured binding host by @Shizoqua in #3668
feat: add explicit custom chunking selector by @er-s-an in #3665
build(deps): update openai requirement from <3.0.0,>=2.0.0 to >=2.0.0,<4.0.0 by @dependabot[bot] in #3670
feat(api): add ENABLE_API_DOCS switch for API documentation exposure by @danielaskdd in #3672
fix(merge): amerge_entities relation weight must track distinct-source count, not max by @AmirF194 in #3674
fix(postgres): make AGE graph initialization idempotent by @danielaskdd in #3677
fix(hf): exclude padding tokens from embedding pooling by @Shizoqua in #3675
fix(kg): rank BFS levels by degree then label on mongo and opensearch by @Dhruv-kys in #3652
Fix chunk-tracking reseed: distinguish absent row from present-but-empty row (#3609) by @jaideeppyne in #3660
fix(graph): enforce relation evidence weight floor by @danielaskdd in #3676
fix(hf): place generation inputs on model device by @Shizoqua in #3678
fix(hf): honor generation token limits by @Shizoqua in #3687
fix(lmdeploy): honor native max_new_tokens by @Shizoqua in #3689
fix(ollama): map max_tokens to num_predict by @Shizoqua in #3690
build(deps): bump lucide-react from 1.29.0 to 1.31.0 in /lightrag_webui in the ui-components group by @dependabot[bot] in #3682
build(deps-dev): bump the build-tools group in /lightrag_webui with 2 updates by @dependabot[bot] in #3683
build(deps): bump the frontend-minor-patch group in /lightrag_webui with 4 updates by @dependabot[bot] in #3684
build(deps): bump react-dropzone from 15.0.0 to 19.1.1 in /lightrag_webui by @dependabot[bot] in #3685
fix(gemini): map max_tokens to max_output_tokens by @AmirF194 in #3696
fix(lollms): map max_tokens to n_predict by @Shizoqua in #3692
test(utils): cover validate_file_path_security and subtract_source_ids by @YashvantHange in #3702
fix(faiss): materialize an overwrite in one atomic rebuild by @danielaskdd in #3705
fix(kg): keep unsaved upserts as a redo log and replay them after a reload by @danielaskdd in #3709
fix(llm): close OpenAI stream on GeneratorExit during error-path COT close by @JessYanCoding in #3713
fix(openai): encode embedding input with disallowed_special=() by @danielaskdd in #3718
fix(embeddings): keep provider API keys request-local by @pengpengyi92 in #3717
chore(deps): allow pandas 3.x by @danielaskdd in #3719
fix(tokenizer): bisect the prefix bracket instead of clamping to one char by @JessYanCoding in #3722
fix(pipeline): keep generated failure summary within varchar(255) by @danielaskdd in #3735
fix(llm): stop retrying permanent spend-stop 429s (LiteLLM budget_exceeded / OpenAI insufficient_quota) by @danielaskdd in #3736
fix(extraction): stop dumping full chunk id list in KEEP skip log by @danielaskdd in #3737
build(deps-dev): bump @types/react-dom from 19.2.4 to 19.2.5 in /lightrag_webui in the react group by @dependabot[bot] in #3730
build(deps): bump lucide-react from 1.31.0 to 1.33.0 in /lightrag_webui in the ui-components group by @dependabot[bot] in #3731
build(deps-dev): bump the build-tools group in /lightrag_webui with 2 updates by @dependabot[bot] in #3732
build(deps): bump mermaid from 11.16.1 to 11.17.0 in /lightrag_webui in the content-rendering group across 1 directory by @dependabot[bot] in #3733
build(deps): bump the frontend-minor-patch group across 1 directory with 4 updates by @dependabot[bot] in #3734
fix(kg): save FaissVectorDBStorage off the event loop by @danielaskdd in #3742
fix: move file-backend persistence off the event loop by @danielaskdd in #3740
fix: keep LoLLMs stream session alive by @gry67673905 in #3741
fix(webui): keep the document list refreshing when the backend stalls by @danielaskdd in #3743
fix: accept valid empty JSON extraction results in cache rebuild by @AmirF194 in #3745
fix(extraction): report an unparseable rebuild payload once, not twice by @danielaskdd in #3749
🐛 fix(webui): stop the conversation area from scrolling horizontally on phones by @danielaskdd in #3750
Fix WebUI document refresh after BFCache restore by @danielaskdd in #3751
fix(webui): show workspace description in tooltip by @danielaskdd in #3769
fix(setup): enable UI templates in compose by @danielaskdd in #3770
fix(setup): close the lightrag service at any top-level compose key by @danielaskdd in #3772
refact(webui): Let the bundle name and render the login consent document by @danielaskdd in #3771
fix(query): enforce meaningful RAG query length by @danielaskdd in #3776
fix(query): refuse empty prompts and stop walking whole queries by @danielaskdd in #3780
fix(reproduce): batch_eval.py always exits with TypeError — give it a CLI by @Anai-Guo in #3779
refact(test): add React component tests and test harness isolation checks by @danielaskdd in #3783
fix: read evaluation datasets as UTF-8 by @zjn20030811 in #3782
Fail legibly when bun test runs without the DOM preload by @danielaskdd in #3784
perf(chunker): defer whole-document encode to the fixed-window path by @hata33 in #3758
refac(test): convert five source-text WebUI tests to real renders by @danielaskdd in #3786
refactor(webui): drop the buildLegacyDocs compatibility layer left over from GET /documents by @danielaskdd in #3791
test(webui): stop replacing the window global in the workspace entry test by @danielaskdd in #3793
fix(export): stop aexport_data crashing on backends without client_storage by @danielaskdd in #3785
perf(operate): encode chunk-invariant extraction prompts once per run by @hata33 in #3760
test(extraction): pin the gleaning guard arithmetic and lazy invariant precompute by @danielaskdd in #3787
docs: clarify file processing concurrency limits by @danielaskdd in #3794
test(webui): finish converting source-text tests to real renders by @danielaskdd in #3792
chore(deps): update anthropic requirement from <1.0.0,>=0.18.0 to >=0.18.0,<2.0.0 by @dependabot[bot] in #3795
chore(deps): widen spacy ceiling to <3.8.17, lock to 3.8.16 by @danielaskdd in #3797
docs(workspace): document the query entry, and show its parameter scope in the WebUI by @danielaskdd in #3798

New Contributors

@El-Tatane made their first contribution in #3629
@Vishwamitra made their first contribution in #3604
@qianheng-aws made their first contribution in #3633
@AmirF194 made their first contribution in #3635
@MasaSN made their first contribution in #3637
@Shizoqua made their first contribution in #3649
@torvalstrom made their first contribution in #3651
@letsgetai made their first contribution in #3473
@ClaireXi99 made their first contribution in #3646
@Marc-oss-hub made their first contribution in #3650
@er-s-an made their first contribution in #3662
@Dhruv-kys made their first contribution in #3652
@jaideeppyne made their first contribution in #3660
@YashvantHange made their first contribution in #3702
@0xPabloxx made their first contribution in #3680
@yzxcj797 made their first contribution in #3693
@JessYanCoding made their first contribution in #3713
@gry67673905 made their first contribution in #3741
@Anai-Guo made their first contribution in #3779
@zjn20030811 made their first contribution in #3782

Full Changelog: v1.5.6...v1.5.7