Repository navigation
- removed the in-repo
npcpy.sqlsubpackage, NPCSQL endpoints, SQL AI adapters, native database AI transformer, NQL/SQL compilation classes, and SQL model compiler; the building-applications guide now redirects users to the standalonenqlCLI (4999b1b, dcc35eb, fa3cc3b, 830cf32, 80e41cc) - removed stale SQL-related example projects, factory files, standalone migration scripts, and the
migrationsfolder (5ce1466, 61d56e1, 05a317d, be9bbed, 1b3829d) - dropped SQL module and
model_runnerjobs from CI and removed obsolete SQL test suites (test_npcsql.py,test_model_runner.py, SQL AI functions tests) (005912b, 5dbb287, 49e83fa, d655a6b, 8c25d4e, dd18d28) - added a
docs-checkCI job that builds the MkDocs site and fails on pages not reachable from navigation (4999b1b, f909e06) - added
standardcomputeanduberllmas named OpenAI-compatible providers with URL/key resolution and env-based or explicit overrides (dcc35eb, d655a6b, 3051ac9, fa50255, 03d82f3, 1b3829d) - added
sofyaas asearch_webprovider (1b3829d, d8bcd0d, 37b4bce, dd18d28) - reorganized the provider directory and README/docs provider sections: alphabetized lists, collapsed sections into
<details>blocks, promoted QLLM-PAM to its own section, added Atlas Cloud, Standard Compute, UberLLM, and an OrcaRouter route under OpenAI, set the Ollama default entry, removed Anthropic and AirLLM entries, and updated the comparison table (4999b1b, 5fb4020, 2f7eab9, 32f4ff1, 5c47a52) - forwarded
seed,frequency_penalty,presence_penalty, andlogit_biasto LiteLLM and ensured they stay single-response (dcc35eb, 0a2cf1c, d28d115) - fixed
top_kforwarding so it is sent only to LiteLLM providers that accept it (anthropic, gemini, ollama, etc.) and omitted for OpenAI (dcc35eb, b2f4424, 49e83fa) - fixed Ollama tool-call streaming to avoid a duplicate
client.chatcall and preserve reported token usage across tool turns and streamed responses (2c26aef, b959c75, d655a6b, af9a697) - fixed
npcpy/streaming.pyto forwardtemperature=0.0in chat and tool-agent streams instead of dropping it (1d54ed9) - fixed the NPC compiler to use
jinxes_directoryconsistently, removing stalenpc_jinxes_directoryreferences, and enabled loading markdown agents with frontmatterjinxes:, H2 agents fromagents.md, and siblingjinxesdirectories for.npcfiles (dcc35eb, 6dc97ab, b57c254) - replaced bare
except:withexcept Exception:innpcpy/gen/embeddings.py,npcpy/npc_compiler.py, and the Jinx MCP conversion (68c7ec9, 6dc97ab, dcc35eb) - added a
system1optional install extra for scikit-learn/sentence-transformers and lazy-loaded heavy ML dependencies sonpcpy.servebinds before spawn timeout (76e44e2, ca52dd4, dcc35eb) - removed the CLI tools section from the README, removed the newsletter signup callout from the docs index, fixed macOS trigger docs to recommend
fswatch, documented thesystem1extra, and updated MkDocs navigation to surface orphaned guides and add a Documentation Index (4999b1b, 91b4ae9, 76e44e2, 6d54706)
What's Changed
- remove deprecated sql module, factory examples; bump to 2.1.28 by @cagostino in #341
- chore: remove example_npc_project folder by @cagostino in #342
- chore: remove migrations folder by @cagostino in #343
- Caug/v2.1.28 by @cagostino in #344
- fix(ollama): avoid double client.chat call when stream=True with tools by @Christian-Sidak in #345
- docs: alphabetize provider list, split QLLM-PAM into its own section by @cagostino in #346
- docs: drop newsletter signup, fix mac trigger dep, surface orphaned guides by @cagostino in #347
- Caug/v2.1.28 by @cagostino in #348
- Add Sofya as a search_web provider by @yusufgurdogan in #349
- feat: add Standard Compute as a named provider by @henrikdukefoss in #350
- feat: add UberLLM as a named provider by @PSA-Team-source in #351
- fix(streaming): forward temperature=0.0 instead of dropping it by @pcbeingused333 in #352
- fix: replace bare except with except Exception in npcpy/data/text.py by @Cmz9111 in #353
- fix: replace bare except with except Exception in npcpy/gen/embeddings.py by @Cmz9111 in #354
- fix: replace bare except with except Exception in npcpy/npc_compiler.py by @Cmz9111 in #355
- fix: forward generation parameters to LiteLLM by @pcbeingused333 in #356
- fix: preserve Ollama usage when processing tool calls by @pcbeingused333 in #357
- fix(litellm): forward top_k to providers that support it by @pcbeingused333 in #358
- fix(npc_compiler): use jinxes_directory, not stale npc_jinxes_directory by @cagostino in #359
- Downgrade version from 2.1.29 to 2.1.28 by @cagostino in #360
New Contributors
- @Christian-Sidak made their first contribution in #345
- @yusufgurdogan made their first contribution in #349
- @henrikdukefoss made their first contribution in #350
- @PSA-Team-source made their first contribution in #351
- @Cmz9111 made their first contribution in #353
Full Changelog: v2.1.27...v2.1.28