Mission

The North Star

Build an agent runtime that stays understandable as agents become more capable.

As models get stronger, the harness around them matters more, not less. A capable model in a tangled, hardcoded runtime is harder to trust than a weaker model in an explicit one. Amber's mission is to keep the runtime legible, so you can understand what the agent is doing, why, and where its boundaries are.

How

Native runtime: compiled C++17 from the agent loop up, not a collection of scripts around an API. One process, no interpreter dependency, the core usable as a library.
Explicit capabilities: providers, tools and UI extensions arrive through the same plugin architecture. The core does not need to know every integration that exists; new ones do not require special cases throughout the engine.
Model / provider separation: the agent runtime is independent of the model provider. Provider integrations are plugins, so the harness works with different vendors and OpenAI-compatible endpoints without changing the agent loop.
Extensibility: plugins can be enabled, disabled and unwound cleanly, with contributions tracked by owner. The architecture is shaped for runtime loading, even if the current set is compiled in.
Measurable behavior: amber-bench runs repeatable scenarios against the same engine and produces KPI scores. Harness changes are tested against the corpus, not judged from demos.
Explicit trust boundaries: model output is treated as untrusted input. Filesystem operations are confined, side-effecting operations can require approval, shell execution is gated. The goal is not to pretend an agent is safe by default, but to make its trust boundaries explicit.
Open source: Apache 2.0, for everyone, without discrimination. Ready for people to try, break, inspect and contribute to.

The Spirit

An interface inspired by BitchX and the golden age of IRC. Fast, keyboard-driven, dense, beautiful. The drawer pops on /, filters as you type, and /q, /sl become muscle memory again. The interface anticipates. It helps. It gets out of your way when you know what you are doing.

Who It Is For

Developers: from a single file to a monorepo
System administrators: server management, automation, deployment
Security professionals: auditing, tooling, long-running investigations
Researchers: data analysis, experiments that run for days
Anyone who lives in a terminal and wants an AI partner

Amber runs on Linux and macOS. It is free and open source (Apache 2.0), for everyone, without discrimination. If you live in a terminal, it is yours.