Local/remote hybrid agents with ultra customizable client harnesses.
Build multi-user agents with TypeScript, Python, or Rust to fit your exact needs.

With Better Agent Engine (BAE), you will:

• Build hybrid agents that execute tools locally OR remotely with durable, replayable sessions.
• Attach mutiple clients to the same session simultaneously for multiplayer collaboration.
• Create ultra-customized agent harnesses to accomplish your specific goals with ease.
• Stay comfortable in your language of choice: TypeScript, Python or Rust.
• Use builtin primitives such as sandboxes, CLI subagents (Claude, Codex, etc.), and MCP servers.

How does it work?

• baesrv is BAE's tiny Rust server that handles LLM provider connections, durable session logs, MCP servers, tool call loops, and auth.
• BAE client libraries (TS, Python, Rust) let you quickly and easily build a custom agent harness by defining tools, sandboxes, prompts, and full lifecycle control of the agent loop with hooks.
• Your harness (local) connects to baesrv (remote) so local tools can collaborate with remote MCP servers and LLM providers to accomplish tasks with the full power of both environments.
• Multiple harness instances can join the same session in parallel to drive prompts, provide tools running in different locations, with each client getting a live stream of the entire session for true multi-user support.
• baectl handles server configuration, operations, and client/profile/session CRUD.

Local and Cloud working together to get complex work done for you and your team.

Status: alpha. The codebase, tooling, and project specification are in place; APIs and SDKs will likely change.

Quickstart

docker run -p 8080:8080 -v bae-data:/var/lib/bae better-agent-engine
curl http://localhost:8080/healthz

Then create a profile and client key via the admin API, exchange the key for a session, and send a message. Full walkthrough: docs/quickstart.md.

Local development

Everything runs in Docker via Make. The only host requirements are docker and make.

make dev-image     # build the dev toolchain image (Rust, Node 22, Python/uv)
make build         # build all four components inside the container
make test          # run all tests
make lint          # clippy / tsc / ruff across components
make fmt           # format all components
make shell         # interactive shell in the dev container

Work on a single component with <verb>-<component>:

make test-server
make build-client-typescript
make lint-client-python

Inside the dev container (or on a host with the toolchains installed) you can also work directly in a component directory: make -C server test.

Run make help for the full target list.

Server

Build the production image and run it with a persistent data volume:

make image
docker run -p 8080:8080 -v bae-data:/var/lib/bae better-agent-engine

Configuration is environment-driven (see aspec/devops/operations.md): BAE_ADDR, BAE_DB_PATH, BAE_LOG, and provider credentials such as ANTHROPIC_API_KEY are passed through the environment, never stored in the database.

Project specification (aspec/)

The aspec/ tree is the source of truth for how this project is designed and operated: foundation, architecture, devops, UX, agents, and work items. New feature work starts as a work item following aspec/work-items/0000-template.md.

License

Apache-2.0 — see LICENSE.