GET STARTED
Install
The fastest path: paste one prompt into your pi agent (or prime-agent). The agent fetches the install guide and does the clone, build, register, restart, and verify for you.
Install with one paste
Install and configure the smithy-agent extension (works under both the pi and prime-agent harnesses) by following the guide here:
https://smithy-agent.netlify.app/install-guide.md Prefer to do it by hand? install-guide.md. Requires Node ≥ 25.
How it works
-
Paste the prompt into your pi agent (or prime-agent).
-
pi fetches the install guide.
-
pi clones, builds, registers the extension, restarts, and verifies.
Prime Agent (dual harness)
smithy also runs under prime-agent, the RLM-native fork of the pi harness — same clone, same build, one extra registration.
-
Add the same absolute
packages/smithy-corepath to theextensionsarray in~/.prime/agent/settings.json. -
Run prime once interactively so its Python kernel runtime is set up, then restart and run
/smithyas usual. -
Verify with
npm run test:live-prime— a live RPC smoke test against the prime binary.
On prime, smithy's subagents (prospector, draftsman, inspector, smith,
judge) adhere to prime's tool model: the persistent
ipython kernel is their only
capability tool — all file, shell, and search I/O flows through it, and
their prompts carry a harness appendix mapping tool names to kernel
equivalents. The full
install guide has the details.
Knowledge base (optional)
For a codebase-aware memory — symbol search, code-context retrieval, a code
graph — also register
smithy-mcp-bridge. It spawns
the tome MCP server and defers
its ~40 tools behind three bridge tools (progressive disclosure). Every
agent can consult the graph when the bridge is online, and falls back to
grep / find / ast_grep when it
isn't. Don't add a separate
mcpServers.tome entry — the
bridge owns it.
- tome_search — find a tool by keyword
- tome_describe — inspect one tool's schema
- tome_call — invoke a tool by name