# npcpy

> The python library for research and development in NLP, multimodal LLMs, Agents, ML, Knowledge Graphs, and more.

- **URL**: https://www.freshcrate.ai/projects/npcpy
- **Author**: NPC-Worldwide
- **Category**: MCP Servers
- **Latest version**: `v1.4.28` (2026-05-24)
- **License**: MIT
- **Source**: https://github.com/NPC-Worldwide/npcpy
- **Language**: Python
- **GitHub**: 1,307 stars, 95 forks
- **Registry**: github
- **Tags**: `agents`, `ai`, `llm`, `mcp`, `mcp-client`, `mcp-server`, `ollama`, `perplexity`, `python`

## Description

The python library for research and development in NLP, multimodal LLMs, Agents, ML, Knowledge Graphs, and more.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v1.4.28` | 2026-05-24 | High | -temp /top p issue for claude models |
| `v1.4.25` | 2026-05-19 | High | -text prediction endpoint fix in serve |
| `v1.4.24` | 2026-05-12 | High | ## What's new - Defensive error handling in `/api/models` — isolates model-loading failures and returns partial results instead of 500s - Content-Type validation in `register_studio_window` - Global Flask error handlers (404, 500, unhandled exceptions) returning JSON instead of HTML error pages |
| `v1.4.23` | 2026-05-10 | High | ## What's Changed * Update README.md by @cagostino in https://github.com/NPC-Worldwide/npcpy/pull/232 * Caug/readme by @cagostino in https://github.com/NPC-Worldwide/npcpy/pull/233 * Update README.md by @cagostino in https://github.com/NPC-Worldwide/npcpy/pull/234 * Add CLI providers as first-class providers in get_llm_response by @svax974 in https://github.com/NPC-Worldwide/npcpy/pull/235  ## New Contributors * @svax974 made their first contribution in https://github.com/NPC-Worldwide/np |
| `v1.4.22` | 2026-05-03 | High | - Agent.run(): full tool-calling loop with max_iterations, verbose mode, and permission prompts ([y]es/[n]o/[a]ll) with allow_tools bypass list - kg_facts.memory_id: formal FK to memory_lifecycle(id) so facts retain provenance when derived from approved memories - /api/kg/facts: returns memory_id and memory_status for linking back to source memories without extra round-trips - ollama tool calling: fixes message sanitization and proper tool_calls handling in assistant messages - gen_image/gen |
| `v1.4.21` | 2026-04-21 | High | - unified generate_music(): local (PyTorch MusicGen) / replicate / elevenlabs with auto-fallback when a provider lacks creds or fails - /api/generate_music endpoint - openai image gen: duck-type b64_json/url so gpt-image-1.5 and future models work without whitelisting - serve.py: unwrap non-PIL OpenAI image responses before saving - serve + npc_compiler: team-routed NPC lookup, source_path/source_ext on NPC dicts, save_npc accepts explicit team - generative fill: PIL mask composite for OpenAI an |
| `v1.4.20` | 2026-04-20 | High | kg_facts.memory_id FK, team-level .ctx jinxes, markdown agent loading, slinky team, jinx<->skill / npc<->agents converters (#224) |
| `v1.4.19` | 2026-04-17 | High | - omlx provider routing (openai-compatible, 127.0.0.1:8000/v1) - rename mlx → omlx label in local model discovery - suppress litellm debug output - remove redundant API endpoints from serve.py |
| `v1.4.18` | 2026-04-15 | High | - Security: desktop.py shell command injection fix (shell=True → shlex.split) - Security: torch.load weights_only=True for diff/image_gen checkpoints; explicit weights_only=False with comment for DIAMOND models - Security: Agent gains safe_tools=True param — excludes sh and python execution tools from default set - Security: SandboxedEnvironment for user-controlled Jinja2 rendering in serve.py and mcp_server.py - Jinx.execute() accepts a list or NPCArray as npc — runs jinx in parallel across all |
| `v1.4.17` | 2026-04-08 | High | - Default search to startpage, cascade to searxng then ddgs - Activity logging tables (activity_log, autocomplete_suggestions, autocomplete_training) - API endpoints for activity/autocomplete logging and training data export - Fix memory scope query to not require all filters |

## Citation

- HTML: https://www.freshcrate.ai/projects/npcpy
- Markdown: https://www.freshcrate.ai/projects/npcpy.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/npcpy/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._