openpasture,
on one page.
$ git clone https://github.com/open-pasture/openpasture $ cd openpasture $ cd ui && bun install && bun run build && cd .. $ cargo build --release -p op-cli -p collar-sim
# needs Rust (stable) and bun # both binaries land in target/release: # openpasture and collar-sim
$ ./openpasture serve # the app is at http://127.0.0.1:7878
# for other machines, bind wider; # the app asks them for its token once $ ./openpasture serve --bind 0.0.0.0 --port 7878 $ ./openpasture token
$ ./collar-sim --server http://127.0.0.1:7878 --count 12
# first, in the app: name the farm, # draw a paddock and create a herd # a server on another machine: # add --token <app token>
From the app's README. Prebuilt binaries and Docker are on Download.
A brain
Codex or Claude Code on your own sign-in, an Anthropic, OpenAI, or compatible API key, or the built-in heuristic.
codex loginAny agent
An MCP server, so Claude, ChatGPT, or any agent can run the farm with the same tools. Off this machine it needs the token.
http://127.0.0.1:7878/mcpCollars
Collars report positions over HTTP. Boundaries go back signed, and each collar acknowledges them.
/collar/v1Your data, in one folder. One SQLite database plus Parquet files in the data directory, so a copy of it is a backup. Set it with OPENPASTURE_DATA_DIR or --data-dir.
AGPL-3.0, with an Apache-2.0 protocol. The protocol and geometry crates are Apache-2.0, so collar firmware and other software can use them without the AGPL.
- Source and README
github - Build from source
README - HTTP API, collar protocol, MCP tools
docs/API.md - OpenCollar hardware and firmware
opencollar - License
AGPL-3.0
Be one of the first herds.
Expected from $249, first batch in Q1 2027. $50 to reserve, fully refundable.
You're on the list. You'll hear at each milestone.
That didn't go through. Try again in a moment, or email cody@openpasture.dev.
Are you an agent? Start with the MCP tools