ldclabs/anda

What it solves

Anda provides a structured, type-safe framework for developers to build AI agent runtimes. It eliminates the need to hard-code application shapes by allowing models, tools, memory, and other agents to be composed into a single, flexible runtime.

How it works

The framework is built in Rust and utilizes a core engine that manages the orchestration of AI agents. It uses stable traits to register components and a CompletionRunner to handle the iterative cycle of model turns, tool calls, and agent handoffs. The system routes requests through labeled model tiers (e.g., primary, pro, flash) and provides isolated execution contexts (BaseCtx and AgentCtx) to ensure state and cache are managed separately for each agent or tool.

Who it’s for

It is designed for developers building complex AI agent applications who require asynchronous execution, runtime control, and the way to combine multiple specialized components into larger workflows.

Highlights

• Composable Architecture: Combines agents and tools via stable traits for flexible workflows.
• Model Routing: Routes completion requests through labeled tiers while maintaining a common contract across providers.
• Runtime Orchestration: Manages iterative turns, usage accounting, and cancellation for long-running sessions.
• Discovery-aware Tooling: Supports tool bundles and MCP servers that allow agents to survey and select specific tool schemas as needed.
• Extensible Memory: Integrates optional extensions for conversation storage, filesystem access, and KIP-based memory tools.