Obsigen AI is a European enterprise conversational agent built to plan, execute, verify, and deliver outcomes — not just generate text. It combines multi-agent orchestration, tool execution (search, file analysis, code sandbox, image generation, PPTX builder), RAG-ready knowledge grounding, and trainability for domain alignment.
With you. Conversation history, uploaded files, generated artefacts and audit logs are written to storage you control — your tenant, your database, your object store, your SIEM. The Obsigen runtime (inference, orchestration) operates on EEA infrastructure, but it doesn't retain your content after the run. On-prem and sovereign-cloud deployments are first-class.
Two things. First, Obsigen is agentic — it doesn't just chat. It searches across sources, analyses your files, runs Python in a secure sandbox, generates images and builds PowerPoint deliverables, with multi-agent Swarm orchestration (50–100 parallel agents) for deep research. Second, your data stays with you — it never trains a third-party model, never leaves your environment, and our runtime doesn't hoard your conversations the way consumer chatbots do.
Obsigen is built for three primary verticals: Manufacturing & Supply (scheduling, maintenance, quality, inventory), Cybersecurity / SOC (alert triage, enrichment, attack surface mapping, DLP), and Sales & Service (pipeline forecasting, creative automation, contact center copilot). The platform is flexible enough for other enterprise use cases as well.
Obsigen has 7 execution strategies that determine how a request is handled: Fast (quick answers), Light (minimal latency), ReAct (tool-based tasks), Master (deep reasoning), Hybrid (tools + lighter swarm), HybridFull (deep research + verify), and Swarm (50–100 parallel agents for maximum coverage). The system can select the strategy automatically or you can choose manually.
Swarm mode deploys 50–100 specialized parallel agents — solvers, critics, researchers, and testers — to tackle complex tasks. The results go through MAP (parallel solving) → REDUCE (consolidation) → VERIFY (stress testing) → FINAL (executive synthesis). This approach significantly reduces blind spots and hallucinations on complex research tasks.
Obsigen includes five built-in tools: WebSearchTool (multi-source search via Brave, SearXNG, Tavily), FileAnalysisTool (PDFs, Office docs, images), CodeExecutionTool (sandboxed Python), ImageGenerationTool (FLUX.1-schnell), and PptDocumentBuilderTool (professional PPTX generation). Enterprise plans can include custom tool integrations.
RAG (Retrieval-Augmented Generation) allows Obsigen to answer questions using your internal knowledge — SOPs, policies, ERP docs, product catalogs, etc. This reduces hallucinations, improves accuracy, and provides traceable citations. Every answer can reference the specific document, section, and version it was based on.
Yes. Options include SFT/Instruction tuning (consistent "house style" answers), LoRA adapters (cost-effective per-tenant specialisation), Preference tuning (alignment to internal standards), and Tool-use tuning (better agent reliability). Training runs on your datasets — either in your on-prem environment, or in our EEA-hosted training cluster under a signed DPA. We never reuse your data to train models for other customers.
Yes — because we hold as little of it as possible. Conversations, files and artefacts persist to your storage, not ours; audit logs flow to your SIEM under your retention. The runtime itself ships with layered controls (TLS, JWT, RBAC, prompt-injection and jailbreak guards, sandboxed code execution with no network egress, MIME-checked file handling, strict CORS, Redis rate limiting, secrets in environment variables). Our operating infrastructure sits in the EEA — but the data of record is yours.
Yes. Code runs in ephemeral Docker containers with no network access, enforced timeouts and per-run isolation. Each execution creates a fresh container that is torn down after completion — nothing persists between runs, and any audit trail of the execution is written to your audit pipeline, not ours.
Obsigen is designed around two principles that make compliance work easier for you: data minimisation (we process, we don't hoard) and customer-held persistence (your storage, your keys, your retention). Inference prompts are transient. No content is forwarded to consumer LLM providers. Audit primitives compatible with EU AI Act Article 12 are emitted in SIEM-friendly format — kept on your side, not ours. We don't claim that Obsigen certifies your compliance — that's your own assessment — but the architecture is built so you can pass it.
Yes. Contact our team to arrange a personalized demo and trial period. We'll set up an environment tailored to your use case so you can evaluate Obsigen with real scenarios.
You can purchase additional message packs as an add-on, or upgrade to a higher tier. We'll notify you as you approach your limit so there are no surprises.
On-premise deployment is available on the Enterprise plan. This gives you full control over the infrastructure, data, and network configuration. Contact our team for details.