| Project | ★ | Merged |
|---|---|---|
| langgenius/dify | 158.6k | 8 |
| bytedance/deer-flow | 83.9k | 3 |
| BerriAI/litellm | 61.3k | 1 |
| labring/FastGPT | 29.8k | 1 |
| 1Panel-dev/MaxKB | 23.0k | 1 |
| HKUDS/DeepCode | 16.7k | 2 |
| neuml/txtai | 13.0k | 1 |
| MervinPraison/PraisonAI | 9.2k | 2 |
| gpustack/gpustack | 5.8k | 1 |
| ApodexAI/FrontierAgent | 5.3k | 1 |
| crmne/ruby_llm | 4.5k | 2 |
| langroid/langroid | 4.1k | 21 |
| modelscope/evalscope | 3.5k | 2 |
| NVIDIA/NeMo-Agent-Toolkit | 2.7k | 1 |
| ant-research/AntOmniEvo | 1.1k | 1 |
| Project | Merged PR |
|---|---|
| dify | #44120 — fix(nodejs-client): replace stale user fields in native upload forms |
| #44097 — fix(nodejs-client): allow empty keyword filters in list APIs | |
| #44090 — fix(nodejs-client): override default headers case-insensitively | |
| #44085 — fix(dify-agent): treat a confirmed-missing Home Snapshot as a success on Enterprise delete | |
| #44084 — fix(api): stage hit-testing query log with flush instead of committing caller session | |
| #44083 — fix(nodejs-client): preserve SSE field value whitespace | |
| #44017 — fix(knowledge): only batch-enable segments whose indexing completed | |
| #44014 — fix(dify-agent): verify E2B binding Home as well as Workspace on acquire | |
| deer-flow | #6966 — refactor(community): fold serper onto the shared max_results |
| #6779 — fix(serper): report a format error for results with no usable objects | |
| #6719 — fix(sofya): report a format error for a results list with no usable objects | |
| litellm | #44800 — fix(anthropic): let /v1/messages mid-stream failures reach the proxy failure boundary |
| FastGPT | #7924 — fix(workflow): 判断器「包含/不包含」对非字符串值恒返回 false |
| MaxKB | #7311 — fix: split at full-width exclamation/question marks in smart_split_paragraph |
| DeepCode | #256 — fix(tools): prune _SKIP_DIRS below the glob base, not the whole path |
| #255 — fix(security): normalize the program path in the destructive-command screen | |
| txtai | #1362 — Honor an empty allow list in the bash agent tool |
| PraisonAI | #5732 — fix(memory): raise on unsupported URL scheme in array form instead of silently using the file backend |
| #5731 — fix: normalise preset name lookup for array-form parameters | |
| gpustack | #6391 — fix(schemas): validate cache provider card icons |
| FrontierAgent | #66 — fix(agent-team): let gate ② track the configurable gate ① for bash / run_python_code |
| ruby_llm | #1052 — Check Mistral batch embedding row positions before pairing vectors with inputs |
| #1051 — Check embedding batch row positions before pairing vectors with inputs | |
| langroid | #1253 — fix(relevance_extractor_agent): send the extraction instructions in the async path too |
| #1250 — fix(chat_agent): route the async strict-recovery retry through the async twin | |
| #1245 — fix(chat_agent): skip strict recovery in the async path when no tool is usable | |
| #1244 — fix(doc_chat_agent): return NO_ANSWER for an empty query in the async path | |
#1243 — fix(openai_gpt): guard a null choices chunk in the async stream path |
|
| #1242 — fix(repo_loader): report a failed clone as None, not as the target path | |
| #1241 — fix(pinecone): raise when add_documents fails instead of only logging | |
| #1238 — fix: allow special-token literals in Parser.truncate_tokens | |
| #1232 — fix: match literal types in Agent output extraction | |
| #1231 — fix: treat empty Redis key deletion as a no-op | |
| #1230 — fix: raise actionable errors for missing document backends | |
| #1229 — fix: retain numbered segment paragraph continuations | |
| #1228 — fix: preserve text across BPE chunk boundaries | |
| #1227 — fix: honor explicit embedding organization configuration | |
| #1226 — fix: deduplicate PDF split boundaries | |
| #1225 — fix: preserve whitespace in line-limited document reads | |
| #1223 — fix: preserve duplicate columns when loading tabular data | |
| #1221 — fix: honor zero overlap in Markdown chunking | |
| #1220 — fix: preserve Markdown hierarchy when heading levels are skipped | |
| #1218 — fix: handle UTF-8 BOMs when loading tabular data | |
| #1215 — fix(llm): handle content-filtered chunks in the async stream path | |
| evalscope | #1833 — fix(messages): absolutise local audio and video paths in markdown output |
| #1832 — fix(multi-choice): accept a full-width comma between unbracketed answer labels | |
| NeMo-Agent-Toolkit | #2345 — fix(observability): resolve the OpenInference span kind from attribute context |
| AntOmniEvo | #20 — fix(proposer): detect degenerate Claude Code sessions before rollout |