Agent SmithAgent Smith is a declarative framework for concurrent AI operatives. It is powered by smithd, a zero-overhead persistent background system daemon, and managed visually via the Agent Smith Workbench desktop IDE.
Agent Smith splits execution and management. Run headless at the edge for production, and inspect visually on your desktop for design.
Persistent System Core
A low-overhead background daemon running over Unix Domain Sockets or serverless edge workers. It acts as the execution router, broadcasting pulses to operatives concurrently, managing SQLite/D1 state, and processing sentinel vetoes in parallel.
Visual Desktop IDE
A cross-platform desktop IDE built in Qt 6 C++ for visual lattice engineering. Map your specialized operatives, write triggers, run real-time pulse traces, and deploy to the cloud in a single click.
Choose the architecture that respects your compute resources. In the war for agency, sequential chaining is corporate compliance; concurrent replication is sovereign autonomy.
A single model hallucination or key error cascades downstream, collapsing the entire task sequence.
Waiting for multiple consecutive API calls. A five-step chain takes 15+ seconds to resolve.
Forcing a single monolithic model to handle formatting, plan layout, SQL audits, and data extraction.
Uncontrolled loops inside black-box frameworks draining enterprise API keys for minor outcomes.
Pulses are broadcast to all specialized operatives concurrently. Response speed is capped by the single fastest turn.
Watcher operatives run passive, lightweight audits on the input/output flow, vetoing or intercepting runs in < 20ms.
State hydration is completely externalized to edge SQLite or Cloudflare D1 databases rather than model context.
Compiled Go binary (`smithd`) or lightweight Worker script initialized in < 1ms to serve edge operations.
How state transitions, broadcast pulses, and watcher interventions operate concurrently within the Lattice.
An external event, webhook, or WebSocket pulse hits the persistent `smithd` daemon.
The daemon queries SQLite or Cloudflare D1 to fetch and hydrate the durable session context.
The pulse is broadcast to the Primary operative and all Watchers simultaneously in memory.
Watchers match regex/schema parameters. Intercepting or vetoing immediately if rules trigger.
The resolved response is logged to memory, and the audited output is emitted back to the client.
Define database connectors, model variables, and concurrent watchers in a single declarative environment.
{
"name": "nebuchadnezzar-deck",
"version": "1.0.0",
"storage": {
"adapter": "sqlite",
"connection_string": "./construct.db"
},
"operatives": [
{
"id": "primary_smith",
"role": "primary",
"system_prompt": "Replicate prompts and execute..."
},
{
"id": "sentinel_sql",
"role": "watcher",
"system_prompt": "Watch for database access...",
"trigger": { "pattern": "select|database|insert" },
"actions": ["intercept"]
},
{
"id": "security_veto",
"role": "watcher",
"system_prompt": "Watch for leak hazards...",
"trigger": { "pattern": "API_KEY|password" },
"actions": ["veto"]
}
]
}Replicate specialized operatives with independent instructions. Run simple models for auditing and larger models for generation.
Apply watcher patterns. Intercept database queries or veto credential leaks instantly before LLM execution compiles.
The exact same `smith.json` configuration runs locally as a native daemon or compiles into a Cloudflare Pages/Worker script.
Execute pulses directly in the browser sandbox. See how the primary operative and concurrent sentinels process commands in real time.
We have completed the WHOIS analysis for the Agent Smith infrastructure. Three elite domains are currently available for purchase and node mapping.
Recommended gateway for developer documentation, user dashboards, and general API credentials.
Perfect endpoint mapping for the daemon registry, local runtime binaries, and CLI repository.
Optimized for inter-agent communications, distributed consensus routing, and telemetry logs.