avalan

Avalan is a multi-backend framework for developing, orchestrating, and deploying AI agents. It is described as supporting use locally, on premises, and in the cloud, and it aims to move projects from prototype to enterprise-scale AI deployments.

It provides a unified CLI and SDK for working with any model and for handling multi-modal inputs. The models section says it can run local checkpoints, on-prem clusters, and vendor APIs through a single CLI or SDK, with named adapters for OpenAI, Anthropic, OpenRouter, Ollama, and other platforms. The same section says models can be swapped between text, vision, and audio, and that the stack can be fine-tuned from tokenizers to reasoning strategies without changing code.

The agents section says avalan can spin up adaptive agents that invoke tools, stream real-time data, and re-plan as context shifts. It also includes built-in memory, described as rich semantic history and pluggable knowledge stores that can contain documents, code, or live web pages. Built-in observability tracks prompts, latency, and rewards, and the tool and flows sections describe wiring in workflows such as PDF and database lookup, database and graph generation, internal microservices, and public APIs, with fine-grained control over authorization, access, and usage scope.