An AI Client You Can Program
A local-first AI client with multi-model support, programmable strategies, and long-term memory.
SUPPORTED PROVIDERS
Architecture
Three primitives for programmable AI systems
Separate orchestration, memory, and model execution.
Host Platform
Local-first host that handles UI rendering, SQLite storage, worker lifecycle, and model provider adapters.
Strategy Engine
Write strategies in TypeScript to control context building, memory routing, and tool orchestration.
Multi-Model Runtime
Run across GPT, Claude, Gemini, and local models without rewriting orchestration logic.
Developer Experience
Code-first strategy control
Define behavior in typed hooks and ship predictable strategy workflows.
export default {
onContextBuild(ctx) {
ctx.slots.add("system", {
role: "system",
content: "You are a helpful assistant."
})
ctx.slots.add("history", ctx.history.recent(10))
ctx.slots.add("input", ctx.input)
return ctx.slots.render()
}
}Strategy SDK
Build reusable strategy modules with explicit lifecycle control.
Type-safe hooks
Compose logic using typed interfaces and predictable contracts.
Memory-aware orchestration
Route context through persistent memory and retrieval.
Context slots
Inject dynamic data into prompts without rewriting strategies.
Open Source
Open. Hackable. Yours.
AfferLab is an extensible platform for programmable AI systems.
