Local AI by design
Connect Agentr to your own Ollama runtime so language-model inference can remain on infrastructure selected by the operator.
CUR-DEV Software · Self-Hosted AI Agent Platform
Create intelligence.
Create intelligence.
Keep control locally.
Agentr helps organizations create, configure, test and publish specialized AI Agents using locally hosted language models and reusable knowledge—while retaining control over deployment, models and business data.
Build · Manage · Embed AI Agents · A CUR-DEV Product
Why Agentr
Specialized AI Agents without giving up infrastructure control.
Agentr separates Agents, model profiles, Knowledge Bases and website publication so each part can be managed deliberately and reused where it makes sense.
Reusable organizational knowledge
Index supported documents into reusable Knowledge Bases and assign the right knowledge to the right Agent without duplicating the underlying content.
Private before public
Create and test Agents inside the authenticated administration area, then explicitly publish only the Agents intended for website visitors.
Different Agents, different jobs
Use separate instructions, model profiles and Knowledge Base assignments for customer support, product information, internal assistance or other specialized roles.
Grounded responses
Retrieve relevant excerpts from assigned Knowledge Bases before generating an answer, with source information and diagnostics available to administrators.
Controlled website publishing
Publish eligible Agents through hosted chat pages or supported iframe integrations with explicit publication settings and website-origin controls.
Inside Agentr
A focused workspace for models, Agents, knowledge and website channels.
Real application screens from the current Agentr development baseline.
Core capabilities
From local models to knowledge-grounded website Agents.
Agents & guided creation
Create multiple Agents with their own identity, instructions, welcome message, suggested questions, model assignment and Knowledge Base assignments.
Local models & Model Profiles
Discover models on the configured Ollama runtime, test connectivity and maintain reusable model settings for different Agent roles.
Knowledge Bases & document indexing
Index supported text, Markdown, CSV, HTML, Word, Excel, PowerPoint and text-based PDF content into PostgreSQL/pgvector for retrieval.
RAG & grounded answers
Retrieve relevant excerpts from the Agent's assigned Knowledge Bases before generating answers, helping keep responses tied to approved organizational information.
Agent Test Lab
Test an Agent privately with its model, behavior and assigned knowledge together, while administrators can inspect retrieval sources and diagnostics.
Website Channels
Publish eligible Agents intentionally, configure a hosted chat experience, define approved framing origins and generate iframe integration code.
Authentication & security foundations
Protected administrator sessions, password controls, TOTP MFA, optional configured email MFA, CSRF defenses, throttling and backend-only model access.
Deployment-ready architecture
Deploy through Docker Compose with a FastAPI backend, PostgreSQL/pgvector, Valkey, background workers, persistent storage and a configured Ollama runtime.
Designed for deliberate publication. Public website Agents should use only knowledge approved for public disclosure. Website-origin controls govern framing behavior; they are not a substitute for protecting confidential content.
Typical workflow
Build an Agent from model and knowledge to website publication.
- Connect the AI runtime.Use the configured Ollama environment and verify the models available to Agentr.
- Create reusable model settings.Define Model Profiles for the language models different Agents should use.
- Build a Knowledge Base.Choose a verified embedding model, upload supported documents and let Agentr index the content.
- Create the Agent.Define its identity, instructions, Model Profile and one or more Knowledge Base assignments.
- Test privately.Use Agent Test Lab to review responses, retrieval sources and eligibility before publication.
- Publish intentionally.Create a Website Channel for an eligible Agent and configure the hosted or iframe experience.
- Maintain the knowledge.Keep sources and Agent behavior aligned with the information the organization wants the Agent to use.
Who it is for
For organizations that want useful AI assistants while keeping control closer to home.
Customer-facing teams
Create public Agents that answer questions from approved product, service or support information.
IT and knowledge teams
Build private Agents around internal procedures, documentation and troubleshooting information.
Technical consultancies & managed-service teams
Operate multiple specialized Agents from one platform while keeping models, knowledge and publication settings distinct.
Organizations prioritizing self-hosted AI
Run language-model inference and business knowledge on infrastructure selected and controlled by the operator.
Architecture
Separate the Agent, model, knowledge and publication layers.
Agentr's architecture is designed so one local model can support multiple Agents, one Knowledge Base can be reused by several Agents, and an Agent can remain active but private until publication is explicitly enabled.
This separation makes it possible to evolve behavior, knowledge and website presentation independently instead of treating every assistant as a one-off chatbot.
AgentIdentity, instructions and behavior
Model ProfileReusable local Ollama model settings
Knowledge BaseIndexed organizational information
Website ChannelExplicit public publication and embedding
Frequently asked questions
Agentr at a glance.
What is Agentr?
Agentr is CUR-DEV's self-hosted platform for creating, managing, testing and publishing specialized AI Agents using local language models and reusable organizational knowledge.
Does Agentr require a cloud AI provider?
No. Agentr is built around a configured Ollama runtime so supported model inference can remain on infrastructure selected by the operator. The deployment still requires appropriate server resources and model availability.
What can I use as knowledge?
The current platform supports text, Markdown, CSV, HTML, DOCX, XLSX, PPTX and text-based PDF content. Scanned PDFs and image-only documents are not currently presented as OCR-supported inputs.
Can different Agents use different knowledge?
Yes. Each Agent can be assigned its own set of reusable Knowledge Bases, while a single Knowledge Base can also be shared by multiple Agents.
Can I publish an Agent on my website?
Agentr includes hosted chat and iframe publication foundations for eligible Agents. Final behavior still depends on the deployment, reverse proxy, browser and framing configuration used for the installed environment.
Is Agentr a multi-tenant customer portal?
Not currently. Administration is designed around authenticated administrators; a complete multi-tenant customer workspace, billing system and delegated customer portal are outside the current product baseline.
Agentr by CUR-DEV
Bring your knowledge and local AI together.
Talk to CUR-DEV about your AI Agent use case, infrastructure, knowledge sources and website integration requirements.