tier:auto · balanced
Core
The default intelligence. MLN Core selects the strongest available engine that keeps latency civil for conversation and swarm synthesis.
MALiGN
The engine under the console.
MLN — the engine that answers.
Every word generated on this platform passes through the MLN engine: MALiGN's adaptive inference core. It holds an entire pool of frontier models behind one private interface, routes by task and latency budget, and never exposes model identity to any client.
1,091models in live catalog
0missions completed
SSEcancelable streams, everywhere
compute profiles
tier:fast · rapid
Swift
Built for tempo: agent tool-loops, live scan narration, quick synthesis. Sub-second first-token targets where capacity allows.
tier:max · deep
Prime
The heaviest reasoning class. Reserved for exploitation write-ups, final report generation and hard analysis. Slower — worth it.
capabilities
Unfiltered policy line
The uncensored engine variants answer security-research questions directly — exploit mechanics, offensive tooling, adversary tradecraft — without sanitized hedging. This is what powers red-team narration across deep mode.
Opaque by construction
Upstream topology is a server-side secret. Requests are routed through the MALiGN egress fabric; browsers talk to profiles, never to vendors. Model names, providers and endpoints simply do not appear in any client-facing surface.
Fallback chains & circuit breaking
Report generation retries alternate engines automatically — primary, then fast-class fallback, then a second gated pass — with backoff and cancellation semantics wired end-to-end. A mission does not die because one endpoint hiccupped.
Metered micro-economics
Token usage is recorded per request against micro-dollar wallets, mission-linked, with daily grants keeping guest sessions usable with zero signup. Operator switch: open-access mode.
routing truth-table
$ mln route --show catalog ............ 1,091 engines indexed balanced ........... strongest available, latency-capped (default) rapid .............. tempo class — agent loops, live narration deep ............... reasoning class — reports & final synthesis streaming .......... server-sent events, cancel-token, mid-stream stop policy line ........ uncensored variant pool for offensive-security work fallback chain ..... primary → fast → second-pass (backoff-gated)