Abliteration AI

Abliteration AI provides hosted, OpenAI-compatible and Anthropic-compatible language-model APIs for teams working with sensitive or high-risk use cases. Its models are positioned for authorized red teaming, cybersecurity research, trust and safety, synthetic-data generation, machine-learning research, and defense or government workflows. Developers can change the API base URL and key in existing clients, then continue using Chat Completions, Responses, or Anthropic Messages request formats. The platform includes streaming, tool calling, vision inputs, and authenticated video inputs on supported paths. A training-data workflow can generate schema-validated datasets such as preference pairs, evaluation rows, classifier examples, and conversation records for model pipelines. Documentation includes quickstarts for Python and Node.js SDKs, Codex CLI, Claude Code, LangChain, LlamaIndex, Vercel AI SDK, Cloudflare Workers, Next.js, FastAPI, Spring Boot, Bun, Cloud Run, and AWS Lambda. For organizations that need controls around an unrestricted model, Policy Gateway adds policy-as-code decisions with allow, refuse, rewrite, redact, or escalate outcomes, reason codes, scoped keys, quotas, and audit-log exports to systems such as Splunk, Datadog, Elastic, or S3. Pricing combines token usage with optional monthly plans: a one-credit free preview is available without a card, while paid plans begin at $20 per month.