live

Overview ¶

Command live runs one agent against a real local model, with in-memory tools and a policy that stops the run for an operator before the counter is changed. It is the shortest complete consumer: an Agent built from render, three tools, a policy, and a terminal tool to finish on.

From a fresh clone, first (once, from the repository root): the consumer modules are pinned to released versions on the module proxy, so running them from the repository root needs a local workspace:

go work init . ./examples/live ./examples/mcp ./mcp ./providers/anthropic ./providers/ollama ./providers/openai

Then:

go run ./examples/live
go run ./cmd/agentrt -db <path> approve <run>
go run ./examples/live -db <path> -resume <run>

The default model is Ollama's own API; -model openai:<id> with -base-url pointing at Ollama's /v1 runs the same thing through the OpenAI-compatible adapter. The database defaults to a fixed name under a per-user cache directory (os.UserCacheDir()/agentrt, created 0700), so repeating the first command above without -db still finds the same file, but another account on a shared machine cannot pre-create or read it the way it could a fixed name in the shared temporary directory.

SIGINT and SIGTERM cancel the run between steps rather than killing it outright: the database is closed in whatever state the last committed step left it, and, once a run has actually been recorded, the exact -resume command is printed on stderr before the process exits non-zero.