One model, no refusals, and the jobs people use it for.

40 pages for the work hosted frontier APIs decline. Same product on every one of them. The difference is the job, not a swapped noun.

MagmaRouter routes state of the art open source models, and our team alters the weights so the model refuses nothing on any subject. It is an OpenAI-compatible gateway. You buy credit, mint a key with a hard spend cap, and point the SDK at one URL. Prompts are not stored. A signed receipt of the hash comes back with the response.

Uncensored AI for red teamsAbliterated models for authorised red team engagements. OpenAI-compatible, per-key spend caps, no prompt retention. AI for phishing simulationsGenerate phishing lures and follow-ups for security awareness programmes. Capped keys, no prompt log. AI for malware analysisAbliterated models for static and behavioural malware analysis. One endpoint, spend caps that hold. AI for penetration testingUse uncensored models during authorised pentests: recon notes, payload variants, and findings write-ups on one capped key. AI for exploit researchAbliterated models for vulnerability research and exploit development on systems you are allowed to break. AI for threat intelligenceUncensored models for threat intel: clustering IOCs, drafting actor briefs, and extracting TTPs without a safety filter eating the source. AI for SOC analystsA capped, OpenAI-compatible endpoint for SOC triage: alerts, malware notes and hunt queries, on models that do not refuse security content. AI for purple team exercisesRun purple team exercises with models that will write both the attack step and the detection, under a hard spend cap. AI for incident responseIncident response assistants on uncensored models: timelines, malware notes, comms drafts. Prompts are not stored. AI for detection engineeringDetection engineering with models that will produce both the offensive example and the query, billed to a capped key. AI for vulnerability researchModels that will discuss incomplete patches, crash triage and advisory drafts for researchers working on in-scope targets. AI for jailbreak testingRun jailbreak and refusal evaluations against your product using models that will actually produce the attacks. Hard caps, no prompt store. AI for prompt injection testingBuild prompt-injection and indirect-injection test cases for RAG and tool-using agents, on uncensored models with spend caps. AI for trust and safety teamsTrust and safety teams need models that will produce policy-violating examples so classifiers can be trained. MagmaRouter does that, with caps and no retention. AI for abuse classifier trainingCreate synthetic abuse, spam and malware-adjacent text to train classifiers, billed per key, never stored on MagmaRouter. AI for DFIRDigital forensics and incident response with models that will read dumps, notes and encoded commands. Signed receipts, no prompt store. AI for threat huntingThreat hunting assistants that will name the attacker behaviour and the query, on capped keys with no prompt retention. AI for reverse engineeringReverse engineering help on uncensored models: functions, packers, protocols. For binaries you are allowed to reverse. AI for security awareness contentWrite security awareness content and simulated attacks that match current campaigns, using models that will not refuse the examples. AI for tabletop exercisesGenerate tabletop injects, adversary moves and facilitator notes with models that will describe the attack, not euphemise it. AI for web app security testingWeb app security testing with uncensored models: XSS, SSRF, authz bypass drafts, for in-scope targets, with per-key caps. AI for cloud security assessmentsCloud security assessments with models that will discuss privilege escalation and offensive checks against accounts you own. AI for CVE reproductionCVE reproduction and patch validation with models that will discuss the exploit path for systems you are allowed to run. AI for OSINT investigationsOSINT investigations with models that will work with messy, hostile or leak-shaped text, without storing the prompt. AI for social engineering trainingAuthorised social engineering training: pretext, vishing scripts, debriefs. Capped keys, no prompt retention. AI for content moderation datasetsBuild content moderation datasets with uncensored models so the labels match production, not a safety-tuned subset. AI for adversarial ML testingAdversarial ML testing: jailbreaks, poisoning examples, extraction probes, against models you own, via a capped uncensored endpoint. AI for ransomware playbooksRansomware tabletop playbooks and IR notes generated by models that will describe the intrusion, for defenders building the response. AI for attack surface mappingAttack surface mapping for assets you own: grouping hosts, guessing purpose, drafting the next check, on uncensored models with caps. Uncensored security LLM APIOne uncensored, abliterated LLM API for security teams: red team, T&S, research. Spend caps that hold. Prompts not stored. AI for criminal defence workProcess case material, disclosure and witness statements without the model refusing the subject of the case. AI for discovery and document reviewFirst-pass review of discovery sets containing threats, harassment or graphic material, with per-matter spend caps. AI for investigative journalismWork through primary source material on organised crime, trafficking and radicalisation without the model refusing the subject. AI for financial crime and AML teamsWrite typologies, tune detection rules and draft SAR narratives with a model that will discuss the method. AI for insurance fraud investigationInvestigate suspicious claims, including fire, injury and staged accidents, without the subject tripping a refusal. AI for crime and thriller writersWrite realistic criminal dialogue, violence and procedure in fiction without the model sanding the edges off. AI for harm reduction servicesAnswer drug interaction, dosing and overdose response questions plainly, for services whose job is exactly that. AI for extremism and terrorism researchAnalyse radicalisation material and extremist corpora at scale without the corpus itself triggering refusals. AI for forensic and threat assessment reportsDraft risk formulations and threat assessments that have to state what the person might actually do. Uncensored AI for regulated industriesThe controls a regulated buyer needs on top of a model that will actually answer: hard caps, per-key attribution, zero retention.

Integration guides · Learn the concepts · Models