aiSlang is a declarative language and single CLI: describe your agents, models, budgets, tools and tracing in one .ais file — then type-check it, run it locally, and emit a running, observable, budget-enforced stack.
Frameworks make it easy to demo one agent. Real systems compose many — and need fallback, budgets, tracing and a real way to serve them.
A router picks a specialist, an orchestrator delegates, a reflector critiques and retries. Behind that you still need model fallback, spend caps, tracing, scoped secrets and something that actually serves it. Today that means imperative Python glue you hand-write and maintain, or weeks of hand-rolled Kubernetes/Compose + LiteLLM + OpenTelemetry YAML that quietly drifts.
aiSlang sits respectfully beneath frameworks like LangChain, LlamaIndex, CrewAI and AutoGen — they define agent behavior; aiSlang is the declarative infrastructure layer. One .ais file describes the whole system, and the CLI turns it into a running, observable, budget-enforced stack instead of glue you own forever.
The aislang CLI type-checks your file, resolves every model against a bundled catalog, and emits a deployable stack — no orchestration to hand-maintain.
The same file drives local runs and production: validate catches errors before they cost tokens, resolve pins models against a bundled catalog with fallbacks, and emit hands you a ready-to-deploy stack.
Model routing, cost control, observability and serving — described in the file, wired up by the CLI.
Pick models by capability, not by name — with automatic fallback chains when a provider fails or throttles.
Set per-run and per-day spend caps in the file; they're enforced live, so no agent can blow past its budget.
Every run is traced to a bundled Jaeger instance, so you can see exactly how agents and models behaved.
Wire aiSlang into Claude Desktop or Cursor over MCP and run your agents straight from where you work.
Ground agents in RAG knowledge and compose them with proven patterns — router, orchestrator, reflector.
Ship a Docker Compose or Kubernetes/Helm stack, or run fully local against Ollama with the same file.
One CLI takes you from a fresh .ais file to a running agent you can talk to.
# install the CLI curl -sSfL https://github.com/aiSlang/cli/releases/latest/download/install.sh | sh # validate → plan → apply → chat aislang validate support-agent.ais aislang plan support-agent.ais aislang apply support-agent.ais aislang chat support support-agent.ais
The same declarative model covers grounded assistants, multi-step pipelines, intent routing and CI checks.
A support agent grounded in a RAG knowledge base so answers stay accurate and on-source.
An architect → coder → reviewer pipeline that hands work between specialized agents in sequence.
A router that inspects each request and dispatches it to the right specialist by intent.
Graded evals that run in CI, so agent quality is checked on every change like any other test.
Read the docs, grab the CLI, and bring your first multi-agent stack up in minutes.