# SimpleLLMFunc

> A simple and well-tailored LLM application framework that enables you to seamlessly integrate LLM capabilities in the most "Code-Centric" manner. LLM As Function, Prompt As Code.               一个简单的恰到

- **URL**: https://www.freshcrate.ai/projects/SimpleLLMFunc
- **Author**: NiJingzhe
- **Category**: Frameworks
- **Latest version**: `v0.8.4` (2026-05-28)
- **License**: MIT
- **Source**: https://github.com/NiJingzhe/SimpleLLMFunc
- **Homepage**: https://simplellmfunc.cn
- **Language**: Python
- **GitHub**: 77 stars, 4 forks
- **Registry**: github
- **Tags**: `agent`, `agent-development-framework`, `agent-framework`, `agentframeworks`, `agentic`, `agentic-ai`, `agents`, `ai`, `llm-agent`, `python`

## Description

A simple and well-tailored LLM application framework that enables you to seamlessly integrate LLM capabilities in the most "Code-Centric" manner. LLM As Function, Prompt As Code.               一个简单的恰到好处LLM应用框架，能够让你以最“Code Center“的方式无缝集成LLM能力。LLM As Function, Prompt As Code

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v0.8.4` | 2026-05-28 | High | ## 0.8.4 - PyRepl Terminal Isolation  ### Improvements - Isolated PyRepl worker stdio at the file-descriptor level so fd 0 no longer inherits the host terminal and fd 1/2 are captured through worker-owned pipes. - Captured direct os.write(...), C extension output, shell command output, and subprocess stdout/stderr into PyRepl stdout/stderr results instead of leaking to TUI hosts. - Detached the worker session on POSIX when possible to reduce access to the host controlling terminal. - Dropped lat |
| `v0.8.2` | 2026-05-26 | High | ## Multimodal Inputs and PyRepl Image Outputs  ### New Features  1. **Multimodal input support** - `llm_function` now supports typed image input through explicit `ImgUrl` / `ImgPath` parameters and lists/unions containing those types. - `llm_chat` now accepts `UserChatMessage` for OpenAI-compatible multimodal user messages containing text and `image_url` content parts. - Added `UserChatMessage.multimodal(...)`, the canonical `UserChatMessage` helper for ergonomic chat message construction.  2. * |
| `v0.8.0` | 2026-05-19 | High | # Change log for SimpleLLMFunc  ## 0.8.0 (2026-05-19) - Callable Decorators, Unified Chat Events, and Provider Defaults  ### ✨ New Features  1. **OpenAI-compatible per-model API defaults**:    - Added optional `api_params` support to `OpenAICompatible` model configs loaded from `provider.json`.    - Instance-level `api_params` are passed to chat and streaming calls.    - Call-level kwargs override `api_params`, enabling stable model defaults with one-off overrides.  ### 💥 Breaking Changes  1. * |
| `v0.7.8` | 2026-04-16 | High | # Change log for SimpleLLMFunc  ## 0.7.8 (2026-04-16) - Responses Adapter and Selfref Fork Context Refinement  ### ✨ New Features  1. **OpenAI Responses API adapter**:    - Added `OpenAIResponsesCompatible` as a first-class interface adapter.    - Supports `provider.json` loading, direct construction with `APIKeyPool`, Responses tool schema translation, and reasoning passthrough.    - Keeps decorator-facing authoring unchanged while mapping system prompts to Responses `instructions`.  2. **Respo |
| `v0.7.7` | 2026-04-03 | High | ## Highlights - fully migrate documentation to Mintlify and remove the legacy Read the Docs / Sphinx pipeline - add bilingual Mintlify docs with Chinese as default and English under `/en` - improve packaged skills and quickstart guidance around skill export, provider organization, strong typing, Pydantic, and Harness Engineering  ## Details - update README and README_ZH release notes for 0.7.7 - align skills, specs, and contributor docs with the Mintlify workflow - clean up obsolete docs trees, |
| `v0.7.6` | 2026-04-02 | Medium | ## Highlights - bundle `simplellmfunc` and `simplellmfunc-developer` Agent Skills with project-grounded docs, examples, provider/env guidance, and prompt-construction notes - add the `simplellmfunc-skill` CLI so installed users can export packaged skills into directories such as `~/.config/opencode/skills` - raise `too_long_to_file` truncation to 20000 tokens and keep full output in temp files with clearer reminders - harden `FileToolset.grep` by rejecting full-wildcard regexes like `.`, `.*`, ` |
| `v0.7.5` | 2026-03-31 | Medium | # SimpleLLMFunc Release Notes (v0.7.5)  ## Overview  `v0.7.5` adds long output truncation support for tools, preventing context overflow from large tool outputs.  ## Highlights  - **Long Output Truncation**: Added `too_long_to_file` option to `@tool` decorator. When enabled, tool outputs exceeding 4000 tokens are automatically saved to a temp file and truncated. - **PyRepl Auto-Truncation**: `execute_code` now has `too_long_to_file=True` enabled by default, preventing context overflow from large |
| `v0.7.4` | 2026-03-26 | Medium | # SimpleLLMFunc Release Notes (v0.7.4)  ## Overview  `v0.7.4` is a correction release that removes the temporary external selfref override path from `PyRepl` and keeps `selfref` on the builtin runtime path only.  ## Highlights  - `PyRepl` now always installs and uses its builtin `selfref` backend. - Removed the temporary `PyRepl(self_reference=...)` override path. - Host-side code should access the builtin selfref backend through `repl.get_runtime_backend("selfref")`. - Targeted PyRepl, runtime |
| `v0.7.2` | 2026-03-18 | Low | ## Highlights - Upgraded the supported OpenAI dependency range to `openai >=2.0.0,<3.0.0` - Switched internal tool-call handling to the concrete function tool-call types used by modern OpenAI SDK releases - Removed reliance on the older `chat_completion_message_tool_call` import path in runtime code and tests - Refreshed package metadata, lockfile, README, README_ZH, and changelog for the `0.7.2` release - Revalidated the full test suite against the upgraded dependency stack ## Breaking Ch |
| `v0.7.1` | 2026-03-17 | Low | # SimpleLLMFunc Release Notes (v0.7.1)  ## Overview v0.7.1 is a runtime polish + tooling release that adds workspace‑scoped file tools, a sandboxed TUI demo, cooperative abort control, and PyRepl runtime improvements. It also stabilizes Langfuse trace context across async tool calls and TUI streaming, fixing split spans in observability. Documentation and navigation were refreshed to align with the updated runtime/tooling contracts.  ## Major Features in v0.7.1 ### 1) FileToolset + Sandbox |

## Dependency audit

- **Score**: 100/100
- **Total deps**: 0
- **Resolved**: 0
- **Unresolved**: 0
- **License conflicts**: 0
- **Warnings**: 0
- **Scanned**: 2026-05-11

## Citation

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

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