mirror of
https://github.com/aaif-goose/goose.git
synced 2026-07-03 14:10:03 +02:00
3e0ce19ddeecb6971044bf6da11a2b4b4ffa3c28
MCP servers (such as gdrive) list out 100s of files so reading each file in the agent loop leads to context limit exceeded errors. we make this change to only read 'active' files, i.e. files that have been viewed by the user using the developer system. It's unclear whether this will work well with other MCP servers cause this requires other MCP servers to use the priority annotation on the resource as well which may be less widely used.
Goose is your on-machine developer agent, working for you, on your terms
GOOSE 1.0!!
This is the branch for goose 1.0 WIP: which is a port over from python to rust + typescript/electron for optional desktop environment. WATCH THIS SPACE
Building
cargo build
Running the CLI
Configure
goose configure
Add/Remove system
goose system add <system_url>
goose system remove <system_url>
OpenAI provider (default):
export OPENAI_API_KEY=...
cargo run --bin goose -- session
Databricks provider (TODO):
export DATABRICKS_HOST=...
export DATABRICKS_TOKEN=...
cargo run --bin goose -- session
Headless mode
Run goose once-off with instructions from a file
Usage: cargo run --bin goose -- run -i instructions.md
GUI
Goose has an electron based GUI which you can see in ui/desktop:
Start sub system server
cd crates/stub-system
cargo run
Troubleshooting
Compiling tokenizers library
tokenizers depends on esaxx-rs which failed to compile because 'cstdint' file
was not found. The following fixed it:
export CXXFLAGS="-isystem $(xcrun --show-sdk-path)/usr/include/c++/v1"
cargo check
Languages
Rust
63.6%
TypeScript
29.6%
JavaScript
1.7%
Shell
1.6%
Python
1.5%
Other
1.8%