Platform

Knowledge Base, Conversational AI, Agent Framework, Advanced Modules, and Analytics — all running in your infrastructure.

Your Organizational Knowledge. Structured, Searchable, and Governed.

The Jarbas AI Knowledge Base ingests documents, images, and audio files in their native formats, processes them through the LLM layer for semantic understanding, and stores them as searchable vector representations in your own database — with no external processing services, no third-party storage, and no data leaving your infrastructure.

Jarbas AI processes all supported content types through the LLM layer directly. The model handles text extraction from scanned documents, semantic understanding of visual content in images, transcription of audio files, and semantic chunking of structured documents — in a single, governed pipeline with no external service dependencies. All ingested content is stored as vector embeddings. Semantic search retrieves passages based on meaning rather than keyword matching. The vector store runs inside your deployment, under your administrator's policies.

Document Formats

PDF, DOCX, PPTX, XLSX, CSV, TXT, MD, RTF. Structured documents are semantically chunked — the platform does not split on fixed character counts, but on semantic boundaries determined by the LLM.

Image Formats

PNG, JPEG, GIF, WebP. Image content is processed by the LLM's vision capability — charts, diagrams, photographs of physical documents, and annotated screenshots are all valid knowledge base content.

Audio Formats

MP3, WAV, OGG, FLAC, AAC. Audio files are transcribed and semantically processed before ingestion. Meeting recordings, voice memos, recorded interviews are all indexable knowledge sources.

Conversational AI

A governed chat interface scoped to the knowledge bases you define. Answers come from your content, attributed to your sources.

The Jarbas AI conversational layer is not a generic chatbot. It is a governed interface where every response is grounded in the knowledge bases you have ingested and the access controls you have configured. Responses cite the specific documents and passages that supported the answer. Users cannot retrieve information from knowledge contexts they are not authorized to access. When knowledge-base-scoped retrieval mode is active, the system does not generate responses from general model knowledge — answers come from your content, attributed to your sources.

Knowledge-Scoped Responses

Every answer is grounded in your ingested documents. No hallucination from general model knowledge when retrieval mode is active.

Source Attribution

Responses cite the specific documents and passages that supported the answer. Traceability is built in.

Access-Governed Retrieval

Users can only retrieve from knowledge bases they are authorized to access. RBAC is enforced at the API layer.

Audit Trail

Every query, every response, and every document access is logged with user identity and timestamp.

Agent Platform

Autonomous AI operations that execute on a schedule, coordinate with each other, and produce structured outputs without manual trigger.

Conversational AI is synchronous — a user asks, the system responds. The Agent Platform extends Jarbas AI with asynchronous, autonomous operations: agents that execute on a schedule, coordinate with each other, and produce structured outputs without requiring a human to initiate each interaction. This is for organizations that have established AI governance through the Inicial deployment and are ready to apply AI to operational workflows.

Autonomous Agent Framework

Define agents with objectives, knowledge base scope, LLM configuration, and execution triggers. Schedule with cron, trigger by events, or invoke from other agents.

Internal Marketplace

A governed repository of agent definitions. Teams create, publish, and share agents through approval and versioning.

Agent-to-Agent Orchestration

Complex workflows decomposed into cooperating agents. Primary agents delegate to specialists, aggregate results, produce composed output.

Execution Audit Trail

Every agent execution produces a complete record: identity, inputs, knowledge queries, LLM completions, outputs, duration, token cost.

Advanced Modules

Advanced Modules extend the Agents tier with industry-specific capabilities — specialized workflows, external integrations, and structured output. See the Plans page for full details.

View Advanced Modules

Operational Intelligence

Visibility into usage, cost, and knowledge gaps. From the platform event log.

Same log as audit trail. No external analytics.

Usage Attribution

By user, department, knowledge base, agent. Chargeback-ready.

Agent Monitoring

Volume, success rate, duration, cost. Failure classification.

Knowledge Gaps

Queries with no results. Questions your content does not answer.

Manager Dashboards

Usage, cost trajectory, coverage, agent health.

Governance That Cannot Be Disabled

Every interaction governed. Every access logged. Every cost attributed.

Audit trail, quotas, RBAC, isolation. Structural.

Immutable Audit Trail

Every interaction recorded. Append-only. User, timestamp, documents, tokens.

Token Quotas

Tracked per user and department. Configurable thresholds. Hard limits.

Adoption and Efficiency Metrics

Incentivize teams to use agents. Measure efficiency, adoption by department, and operational ROI.

Data Masking

Pattern-based rules. Prevent sensitive data in responses.

Knowledge Base Isolation

Module-scoped. Validated by automated tests on every build.

RBAC

Explicit role assignment. Least privilege. No default grants.

Security Is the Architecture, Not a Configuration

Structural controls. Cannot be disabled or overlooked.

Parameterized queries, JWT, magic bytes, audit. Built in.

Data Isolation

Single-tenant. Database, API, worker, vector store — all dedicated to you.

Authentication

JWT with exp, iss, aud. RBAC at API layer.

Audit Logging

Every interaction logged. Immutable. No opt-out.

Network Security

CORS from configured origins only. HTTPS enforced.

Your Infrastructure. Your Data.

AWS, Azure, GCP, or on-premises. Dedicated. Isolated. Your choice.

Deploys where you operate. No data migration.

Cloud (Your Account)

AWS, Azure, GCP. Your IAM, your security groups. JBTech has no standing access.

On-Premises

Bare metal, VMware, Hyper-V, OpenStack. Data residency compliant.

Hybrid

LLM calls to on-premises endpoints. App tier in cloud.

JBTech-Managed

Single-tenant. Isolated. Available for AWS, Azure, GCP.

AWS

Azure

GCP

The Model Is a Parameter. Not a Commitment.

Choose at deployment. Change without touching the application.

Unified abstraction. Switch providers — not code.

OpenAI

General-purpose. Default connector.

Anthropic Claude

Analytical tasks. Long-context. Knowledge retrieval.

Google Gemini

High-throughput. Cost-effective for agents.

Snowflake Cortex

Vector search and AI inference on Snowflake.

On-Premises

Ollama, vLLM, or custom endpoints.