King AI Orchestrator

King is an orchestrator-worker harness: your agent is the King who decomposes, routes, verifies, and synthesizes - while a court of free-tier LLM providers does the token-generation grunt work. Instead of forwarding one prompt to one model, King breaks a task apart, sends each piece to the best-fit worker, fans out the correctness-critical spans for a consensus check, and merges everything into a final artifact it actually stands behind.

The court has three ranks:

Peasants lift - bulk translation, drafting, classification, transcription, code.

Advisors judge - they critique drafts, adjudicate disagreements, and review risk, but they advise; the King decides.

And the King - your agent - owns the plan and the final answer.

Best for: Big jobs it can even field teams: parallel sub-agents, each running the full loop over its own workstream.

Everything routes by intent. You ask for a translation, a code fix, or a deep-reasoning pass; King maps it to the right lane and the right worker automatically, with automatic failover when a provider rate-limits or errors - so the work just keeps flowing. It's built to be lazy but never negligent: spend the minimum tokens that still hit the quality bar, and never cut verification to save them.

Privacy is a hard wall. Confidential work is routed only to no-training providers - starting with your own on-device local model, so nothing ever leaves your machine. The engine refuses to violate this, by design. Best For : King works brilliantly whenever something needs to be big, tedious, or gets a second, third, and fourth look - the places where the single agent is slow, too costly, or just too naive to bet on. Heavy lifting tasks Bulk translation (entire documents and multilingual batches), spotting discrepancies in contested terms via an expert-checked lane; Bulk classification, tagging, extraction; Audio transcription; Draftings in bulk and variation, quick initial passes, and boilerplates. The areas where the decision matters Cross-checking facts; Critique & risk analysis of drafts, plans, or decisions; Complex reasoning and in-depth analyses across multiple stages.

Construction and investigation tasks Code generation and debugging (leveraging specialized code models and fast generalists); Research and synthesis (collecting and summarizing information from multiple sources); Large and complex multi-stage projects (distributed among smaller teams and reconciled). Confidential content Sensitive data, client files, and legal documents are routed to an on-device, private model, ensuring they never leave the system The King ofdelegation +verification: It shines attasks you'd dorepeatedly, where a costly mistake is possible witha lone agent, and whereyou get a great benefitfrom delegating time-intensive work to several models while your own agentmaintains strategic control and verifiesthe final output.

Not ideal for: single, simple requests where calling the entire court is a burden; hard-real-time applications where additional delays may be detrimental. King prioritizes quality and efficiency over extreme low-latency.

Getting started is now a guided flow. The built-in onboarding wizard walks you through everything on first run:

Collect your keys your way - paste them one at a time with step-by-step instructions for each provider, or bulk-import them from an Excel/CSV sheet.

Set up a private local lane - it detects your local model runner, helps you pick a model, and wires it in for zero-egress confidential work.

Add any provider you like - point King at a new service and it scans the model's capabilities, figures out what it's good at, and slots it into the routing matrix for you - safely, as a fallback lane, without disturbing anything already working.

Paid escalation exists but is strictly opt-in: King only reaches for a premium model after the free workers genuinely fall short - and it always tells you when it spent.

The intelligence lives in the instructions, not a black box. King's Python is a set of small, reliable, transparent primitives; the strategy lives in a human-readable playbook your agent follows. You can read it, edit the routing, and see exactly what happened - every call is logged so you can watch the free tiers doing the work your brain didn't have to.

Be the king.

(Use and abuse of free inference at own risk)