Build long-horizonreliableparallel agents
in Rust.

Appam is a Rust framework for engineering AI agents in production environments where concurrency, durability, traceability, and long-running orchestration matter. It is built for high-availability, high-throughput agentic jobs with built-in tracing, session persistence, configurable continuation mechanics, and provider-aware runtime controls.

Appam is designed for production workloads with automatic failure handling, provider-specific caching support, and durable traces that can be queried after the run.

What you get

• Production-oriented runtime — engineered for long-horizon, multi-turn, tool-using workloads rather than single-shot demos.
• Nine provider variants through one API — switch providers with a one-line change.
• Typed tool system — define tools as Rust structs, closures, or TOML declarations.
• Streaming by default — real-time events to console, channels, callbacks, or custom consumers.
• Traceability built in — emit JSONL traces, stream structured events, and inspect runs after the fact.
• Session persistence — conversations survive restarts via SQLite and can be resumed or queried later.
• Reliability controls — retries, continuation mechanics, rate limiting, and provider-specific tuning knobs.
• Cost-aware execution — use provider routing and caching features where supported.

Production optics

Appam is designed for teams that need to run many agent sessions concurrently without losing visibility into what happened during each run. The core posture is:

• High throughput through async Rust and concurrent provider clients.
• High availability through retry handling, continuation logic, and resilient streaming paths.
• Long-horizon execution through persisted session history and resumable agent loops.
• Traceable execution through structured stream events, JSONL traces, and SQLite-backed history.
• Easy extensibility through Rust-first tools, closure tools, TOML agents, and optional Python bridges.

Supported providers