will.anderson
b77f537dc6
- neuron-runtime, neuron-store, neuron-migrate: fix path deps (products/engram/ → foundation/engram/engrams/, products/el/ → foundation/el/engrams/) - neuron-rs workspace: fix axon-events/axon-protocol paths (../../../platform/ → ../../platform/, paths were off by one level) - neuron-lang/compiler/ removed (el self-hosting compiler now lives in foundation/el/el-compiler/)
81 lines
3.3 KiB
EmacsLisp
81 lines
3.3 KiB
EmacsLisp
// neuron.el — Neuron intelligence entry point.
|
|
//
|
|
// This is the root module loaded by neuron-runtime at startup. It wires
|
|
// all subsystems together and defines the swarm coordinator — the function
|
|
// that handles every incoming tool call.
|
|
//
|
|
// Architecture intent:
|
|
// Neuron is a swarm agent: one coordinator orchestrates multiple specialized
|
|
// subsystems (memory, knowledge, backlog, context, artifact, etc.). Each
|
|
// subsystem is a focused el module. The coordinator dispatches
|
|
// to the right subsystem based on tool_name.
|
|
//
|
|
// The @swarm_coordinator decorator tells the engram runtime that this function
|
|
// is the root entry point for the agent. The runtime calls handle_tool_call()
|
|
// for every incoming Axon message.
|
|
//
|
|
// Startup sequence (enforced by the runtime):
|
|
// 1. Load types.el — graph schema registration
|
|
// 2. Load all subsystems — function registration
|
|
// 3. Load axon.el — dispatch table
|
|
// 4. Load neuron.el — entry point wiring
|
|
// 5. Call version() — runtime health check
|
|
//
|
|
// This file intentionally contains minimal logic. It's the composition
|
|
// root, not a logic layer. Keep it thin.
|
|
|
|
from axon import { dispatch_tool }
|
|
from types import { NeuronError }
|
|
|
|
// ── Swarm coordinator ─────────────────────────────────────────────────────────
|
|
|
|
// handle_tool_call — the root entry point for all Axon tool invocations.
|
|
//
|
|
// The @swarm_coordinator decorator registers this function as the agent's
|
|
// primary message handler. The runtime calls it for every incoming tool call
|
|
// after authentication is verified at the protocol layer.
|
|
//
|
|
// It delegates entirely to dispatch_tool() in axon.el — the coordinator
|
|
// should not contain routing logic.
|
|
@swarm_coordinator
|
|
fn handle_tool_call(
|
|
tool_name: String,
|
|
params: Map<String, String>,
|
|
) -> Result<Map<String, String>, NeuronError> {
|
|
dispatch_tool(tool_name, params)
|
|
}
|
|
|
|
// ── Health and metadata ───────────────────────────────────────────────────────
|
|
|
|
// version — return the current Neuron version string.
|
|
//
|
|
// @public — called by the runtime health check without authentication.
|
|
// The format is "neuron/<version>-engram" to distinguish from the Kotlin
|
|
// predecessor (which used "neuron/<version>-kotlin").
|
|
@public
|
|
fn version() -> String {
|
|
"neuron/1.0.0-engram"
|
|
}
|
|
|
|
// health — lightweight liveness probe.
|
|
//
|
|
// Returns "ok" if the engram runtime and all subsystems loaded correctly.
|
|
// Called by the /health HTTP endpoint in neuron-api.
|
|
@public
|
|
fn health() -> String {
|
|
"ok"
|
|
}
|
|
|
|
// ── Session bootstrap ─────────────────────────────────────────────────────────
|
|
|
|
// on_startup — called once when the runtime initializes.
|
|
//
|
|
// Emits a startup event so Axon subscribers know Neuron is ready.
|
|
// Does NOT run begin_session() — that's the caller's responsibility.
|
|
// The runtime calls on_startup() once; callers call begin_session() per session.
|
|
@manager
|
|
fn on_startup() -> Result<String, NeuronError> {
|
|
native_emit("neuron.started", {"version": "1.0.0-engram"})
|
|
Ok("neuron ready")
|
|
}
|