Rig agent platform: release 0.43.0 with local YOLOv8 pose estimation
On September 30, 2026, release 0.43.0 of the Rig platform came out — a Rust library for building modular and scalable LLM-based applications.
What's new
- Local YOLOv8 pose estimation — rig-candle now supports pose estimation, which runs on your own hardware without contacting cloud providers.
- Automatic explicit caching in Gemini — caching is enabled for long-running jobs, reducing repeated processing costs.
- Unified provider vocabulary and lossless configuration — changes in core and ecs make it easier to move configurations between different providers without data loss.
- typesafeai provider — initial support for a new provider has been added.
- Usage tracking as Option — usage counters can now be absent, which makes the data model more accurate for providers that do not return such information.
- Improvements in rig-ecs — the ECS contract was verified in a long-running cycle of tool operation across five threads, tool results are represented as data, cached representations of statements and explicit stream delivery have been added.
What the platform does
Rig is a Rust library for building agentic applications with support for multi-turn streaming and prompting. The documentation lists its own tools and functions, agent memory, a knowledge base via 10+ integrations with vector stores, as well as support for transcription, audio and image generation.
Your own model instead of the cloud
Rig connects local models via Ollama, an OpenAI-compatible API and llama.cpp. It is installed via cargo add rig or cargo add rig-core. License: MIT.
How it differs from its neighbors
Unlike CrewAI, Agno and AgentScope, which are written in Python, Rig is a Rust library, which gives access to the cargo ecosystem and compilable binaries. According to the feature comparison, Rig and its neighbors share their own tools, agent memory and knowledge base, but multiple agents in a pipeline, MCP support, the A2A protocol and human approval are not stated in Rig's documentation — most of its neighbors mention them.
| Rig | CrewAI | Agno | AgentScope | |
|---|---|---|---|---|
| Type | framework | framework | framework | framework |
| Language | Rust | Python | Python | Python |
| License | MIT | MIT | Apache-2.0 | Apache-2.0 |
| GitHub stars | 8,782 | 58,728 | 42,233 | 31,923 |
| Releases in 90 days | 4 | 23 | 29 | 6 |
| Multiple agents in a pipeline | — | yes | — | yes |
| Own tools and functions | yes | yes | yes | yes |
| MCP support | — | yes | yes | yes |
| A2A protocol | — | yes | yes | yes |
| Agent memory | yes | yes | yes | yes |
| Knowledge base (RAG) | yes | yes | yes | yes |
| Sandboxed code execution | — | — | — | yes |
| Human approval | — | yes | yes | — |
| Graphs and state persistence | yes | yes | — | — |
| Tracing and debugging | — | — | yes | — |
| Voice and speech | yes | — | — | yes |
| Own server and Docker | — | — | yes | — |
| Images and documents | yes | — | — | yes |
| Own model: Ollama | yes | yes | yes | yes |
| Own model: llama.cpp | yes | — | — | — |
| Own model: LM Studio | — | yes | — | — |
| Own model: OpenAI-compat. API | yes | — | yes | — |
Comparison based on the projects' READMEs and documentation as of 10/01/2026: "yes" means the feature is stated there, a dash means it is not mentioned (this does not mean the feature is absent). Stars and releases are according to GitHub data.
It is worth a closer look for those who already write in Rust or want to build agents with local inference via Ollama and llama.cpp. It is convenient to follow such platforms in the CPU3D news feed.
