Epsilla
Epsilla is an Agent-as-a-Service platform that lets enterprise teams build vertical AI agents without writing code, backed by managed RAG and its own vector database. The free tier is permanent; paid plans open at 29 USD per month.
What is Epsilla?
Epsilla presents itself as Vertical AI Agent as a Service, and the homepage promise is blunt: build vertical AI agents without engineering overhead. The bet is that the domain expert, not the infrastructure team, should be the one shipping the agent.
Six platform benefits are announced: agent and chatbot management, a no-code AI agent builder, RAG as a Service, scalable infrastructure, enterprise-grade multi-tenancy and flexible deployment options. Two distinct products sit underneath. Epsilla Cloud is the agent platform, where knowledge bases are assembled and agents configured and published. Epsilla Vector Database is the retrieval engine, which takes a parallel graph traversal approach instead of the usual HNSW index.
Knowledge arrives through ten documented source types: local files, websites, Google Drive, Amazon S3, Notion, SharePoint, Google Cloud Storage, Azure Blob Storage, Confluence and Jira. Models are connected with your own API key, from OpenAI, Anthropic and Mistral to the embedding providers Jina AI, Voyage AI, Mixedbread AI and Nomic AI, while Tavily handles real-time web search. Deployment goes three ways: managed SaaS, on-premise installation, or dedicated private cloud.
Eight verticals are advertised, namely manufacturing, research management, inbound sales and customer support, construction, publishing and libraries, financial services, healthcare and legal services, and the documentation addresses three audiences: domain professionals, GenAI builders and enterprises.
The publisher is Epsilla, Inc., a Delaware company headquartered at 605 Pavonia Ave, Jersey City, New Jersey. It went through Y Combinator's Summer 2023 batch, where it is still listed as active in the San Francisco Bay Area. Its three co-founders all come from TigerGraph: Richard Song, CEO and former Senior Director of Cloud Engineering, Ricki Qin, former Engineering Manager for Cloud, and Eric Yang, former architect of TigerGraph Cloud.
The maturity signals are quiet but real: a domain registered in April 2023, a first web archive capture on 11 August 2023, GitBook documentation on docs.epsilla.com, a console on cloud.epsilla.com, a Discord community, a GitHub organization, a YouTube channel, Y Combinator, NVIDIA, IDC and Product Hunt badges, and eight named testimonials from organizations including TigerGraph, Stanford School of Medicine and Seoul National University. A permanent banner also promotes HarnessRouter, a sibling product offering a single API to run agents such as Codex, Claude Code and Hermes inside your own software.
What it does
- Build AI agents without writing code, through drag-and-drop and a visual flow editor
- Connect a private knowledge base and question your own documents in natural language (RAG)
- Publish and share an agent, or embed it in your own product through the API
- Manage, monitor and optimize the performance of agents and chatbots
- Deploy as managed SaaS, on-premise or in a dedicated private cloud
- Run several agents in enterprise-grade multi-tenancy with granular permissions
When to use Epsilla / When not to
A quick filter to help you decide if Epsilla is the right fit.
When to use Epsilla
- Non-technical business teams that want an agent grounded in their own private knowledge base, with an API left available for whoever plugs it into a custom interface later
- Enterprises with data residency constraints: on-premise installation and dedicated private cloud sit alongside the managed SaaS
- Enterprise RAG projects that need to ingest from Google Drive, Amazon S3, Notion, SharePoint, Confluence and Jira
- Software vendors after a white-labeled chatbot, with extended white-labeling on the Enterprise Tier and the 'Powered By Epsilla' mention removable as a 39 USD/month add-on
- Research and knowledge-heavy organizations: Stanford School of Medicine reports managing research publications on Epsilla after a handful of setup steps
When not to use Epsilla
- Anyone handling protected health information: section 4.5 of the terms forbids using the software to store, process or transmit PHI without a Business Associate Agreement
- High Risk Activities, explicitly excluded by section 3 of the terms: nuclear facilities, air traffic control and life support systems
- Teams that intend to benchmark the service or build a substantially similar product, both contractually prohibited by section 3
- Solo users hoping the free plan will carry a real workload: one member, one application, 50 messages a month, and data cleared after 30 days of inactivity
- Mobile-first users and non-English-speaking teams: no iOS or Android app exists and no interface language other than English is declared
How to use Epsilla
A typical end-to-end flow, from setup to results.
- Sign up and log in to Epsilla Cloud at cloud.epsilla.com; the free plan asks for no credit card
- Create a knowledge base from your local files or from a cloud storage connector
- Adjust that knowledge base's loading, parsing and embedding settings if the defaults do not fit your documents
- Inspect the data storage view to check what was actually indexed
- Add a model provider by entering your own API key and having it validated; the models then become available across the platform
- Build the agent: define its role, attach the knowledge bases, then set its appearance and interaction style
- Review quality in the Evaluation section, which keeps a history of evaluation runs
- Publish and share the agent, or call it through the API to power your own interface
- Manage team members, projects and billing from the console; knowledge bases can also be driven programmatically
- Ask for help at info@epsilla.com, on Discord or through the 'Help Me' chatbox; on-premise and private cloud deployments go through the sales team
Pros & Cons
Pros
- A permanent free plan with no credit card, so the product can genuinely be evaluated before any commitment
- Three deployment modes including on-premise and dedicated private cloud, which is rare at this price level
- Numerous documented enterprise connectors: SharePoint, Confluence, Jira, Amazon S3, Azure Blob Storage, Google Cloud Storage and more
- You can bring your own LLM keys, which bypasses the monthly message quotas entirely
- GDPR commitments written into the contract as an Article 28(3) processing agreement rather than a bare compliance claim, with customer ownership of Customer Data explicitly acknowledged
- A public, quantified SLA with a service credit schedule
- Credible backing: founders out of TigerGraph, a Y Combinator batch, detailed public documentation, a Discord community, and eight named testimonials including Stanford School of Medicine
Cons
- No readable security certification: the Vanta trust center exists but its content is rendered entirely in JavaScript, so nothing verifiable appears on the page
- No sub-processor list is published; AWS is the only third party named, and only inside the privacy policy
- Nothing about training models on customer data, and no documented opt-out
- No hosting country and no hosting region is specified anywhere
- Tight message quotas outside Enterprise: 50 a month on the free plan, 500 on Starter, 5,000 on Professional, with free-plan data cleared after 30 days of inactivity
- Hard commercial terms: non-refundable and non-cancellable fees, automatic one-year renewal with 60 days' notice, and liability capped at the greater of the last twelve months' fees or 10,000 USD
- Thin elsewhere: a generic privacy policy in force since 16 November 2023 and written for a website more than an agent platform, no mobile app, no browser extension, no Article 27 EU representative, no named DPO, and no declared interface language
Pricing & Plans
Epsilla operates on a freemium basis. A permanent Free Tier is offered at 0 USD per month with no credit card required; this is a standing free plan, not a time-limited trial. The lowest paid entry point is the Starter Tier at 29.00 USD per month, billed monthly. Above it, the Professional Tier is 249 USD per month and AI Concierge 2,499 USD per month, while the Enterprise Tier is quoted on request with customizable billing cycles. The pricing page offers a monthly/annual toggle. Six add-ons are billed separately: 12 USD per seat per month for an extra team member, 2 USD per month per AI application, 2 USD per month per knowledge base, 18 USD per GB per month of vector storage, 3 USD per 100 messages per month, and 39 USD per month to remove the 'Powered By Epsilla' mention. All fees are expressed in United States dollars and are non-refundable and non-cancellable under section 2.1 of the terms; overdue balances carry a 1.5% monthly late charge under section 2.3.
- 1 member
- 1 project
- 1 AI application
- 1 knowledge base
- 10M of vector storage
- 50 messages per month (GPT-4.1)
- community support
- no credit card required
- 2 members
- 1 project
- 5 AI applications
- 5 knowledge bases
- 1 GB of vector storage
- 500 messages per month (GPT-4.1)
- email support
- 20 members
- 10 projects
- 50 AI applications
- 50 knowledge bases
- 10 GB of vector storage
- 5
- 000 messages per month (GPT-4.1)
- dedicated support channel
- everything in Professional
- plus a fractional Epsilla AI engineer who builds the agents and workflows for you
- business domain upskilling
- weekly strategy check-ins
- custom integration support
- a priority support channel
- continuous optimization
- and the option to return to a self-serve plan at any time
- unlimited members
- projects
- AI applications
- knowledge bases
- vector storage and messages
- any LLM provider
- extended white-labeling
- SSO (SAML
Data, GDPR & hosting
A consolidated view of how Epsilla handles your data.
GDPR overview
GDPR is handled contractually rather than through a badge. Section 4.6 of the terms provides that where Customer Data contains personal data under Regulation 2016/679, the agreement itself constitutes a contract within the meaning of Article 28(3): the customer acts as controller, Epsilla as processor. Eight processor commitments are listed, from documented instructions, staff confidentiality and Article 32 security measures to the Article 28(2) and (4) conditions for engaging a sub-processor, assistance with data subject requests under Chapter III, assistance with Articles 32 to 36, deletion or return of the personal data once the service ends, and making information available for audits. The silences count too: no Article 27 EU representative, no named DPO, requests routed to one generic address, and no international transfer mechanism. Terms updated 17 April 2025; privacy policy effective 16 November 2023.
Who owns the data?
Section 4.4 of the terms of service is unambiguous: all title and intellectual property rights in Customer Data belong exclusively to the customer. The customer remains solely responsible for what those data contain, must collect consent from its own end users, and undertakes not to expose sensitive or regulated data to Epsilla in unencrypted form. In exchange, the customer authorizes Epsilla to access, process and store the data for the sole purpose of delivering the service. One asymmetry deserves attention: under section 4.3, any suggestion or feedback sent to Epsilla is irrevocably assigned to Epsilla. Terms last updated 17 April 2025.
Reuse rights
Section 6.1 allows Epsilla to use Customer Data for internal business purposes only, which it defines as providing support and improving, testing and delivering the services. The wording is broad and covers more than day-to-day operation. Derived usage data escapes the confidentiality obligations once anonymized and aggregated, and anything submitted through a support ticket is likewise excluded from the definition of confidential information. The privacy policy, effective 16 November 2023, states that Epsilla does not sell, rent or lease information to third parties, while allowing sharing with trusted partners for statistical analysis, email and postal mailings, customer support and deliveries; those partners may use the data only to perform those services. Pages visited are tracked to identify the most popular services and to personalize content and advertising. Disclosure without prior notice is limited to cases required by law or needed to protect rights and safety. Nothing is said about training models on customer data, and no opt-out is documented: the subject is simply never addressed.
Data retention & training
Hosting summary
The privacy policy names a single infrastructure provider: Amazon Web Services, used to run Epsilla's cloud-based retrieval augmented generation service, with an explicit pointer to the AWS data protection page. Beyond that, hosting is undeclared. No country, no region and no data residency commitment appears on any of the collected pages. Data in transit is protected by SSL encryption and the site states that credit card numbers are not stored. For organizations that must control location, Epsilla's answer is architectural rather than contractual: on-premise installation and dedicated private cloud are both offered and are presented precisely as the response to data residency requirements. One technical clue, and no more than a clue: the domain resolves to 76.76.21.21, an anycast node geolocated in the United States on AS16509 (Amazon.com). That describes the edge infrastructure serving the website, not where customer data lives. The publisher is a Delaware company headquartered in Jersey City, New Jersey, so United States jurisdiction should be the working assumption until Epsilla confirms otherwise in writing.
Things to keep in mind
Risks and trade-offs to weigh before adopting Epsilla.
- The security posture cannot be verified from the site: a Vanta trust center is advertised but its content is entirely JavaScript-rendered, so no SOC 2 or ISO 27001 attestation is readable
- No sub-processor list is published, AWS being the only third party named, and nothing states whether customer data is used to train models or how to opt out
- Section 6.1 lets Epsilla use Customer Data to improve and test the services: broad wording that is worth narrowing in writing before uploading anything sensitive
- Protected health information is forbidden without a Business Associate Agreement (section 4.5), and free-plan data is deleted after 30 days of inactivity, an easy way to lose work you assumed was safe
- Commercial lock-in: fees are non-refundable, renewal is automatic for one-year periods with 60 days' notice, while Epsilla may terminate for any reason on 30 days' notice (section 5.3)
- A bursting clause allows chargeable overages, capped at 50% of the volume and 30 days, and liability is capped at the greater of the last twelve months' fees or 10,000 USD
- Governing law is Delaware, United States, with no European forum, and Epsilla may name the customer as a commercial reference (section 6.3)
Setup & Integrations
Technical difficulty
Low on the standard path. Sign-up is open, the free plan needs no credit card, and the documented journey has three steps, build a knowledge base, configure the agent, publish, with no code required; the documentation claims a working agent in minutes, and Pyramid AI reports a specialized chatbot live in minutes. The main friction is validating an LLM provider API key, where the documentation admits OpenAI's tiered billing can confuse. Advanced use, the API, programmatic knowledge bases, fine parsing and embedding settings, asks more, and on-premise or private cloud deployment leaves the self-serve path entirely.
Deployment
Integrations
Behind Epsilla
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Epsilla.
Frequently asked questions
Do I need to know how to code to use Epsilla?
Is there a free plan?
How much does the cheapest paid plan cost?
Which data sources can I connect?
Which AI models can I use?
Can Epsilla run on our own infrastructure?
Who owns the data I upload?
Is Epsilla GDPR compliant?
Is there an uptime commitment?
Can I process healthcare data?
Should you pick Epsilla?
Epsilla is a coherent, well-documented product built by a team out of TigerGraph that passed through Y Combinator, and it knows exactly who it is talking to: the domain expert who has the knowledge but not the engineering budget.
Its real differentiator is not the no-code builder, since everyone claims one, but the pairing of deployment freedom with connector depth. On-premise installation and dedicated private cloud are rarely available at this price level, and the ten documented knowledge sources, SharePoint, Confluence, Jira and the major object stores included, describe a product that has gone to meet enterprise data where it already lives. Bringing your own LLM key, which sidesteps the message quotas altogether, is another sign of a platform built for people who will outgrow the defaults.
The weak spot is just as clear: transparency. There is no readable security certification, the trust center rendering nothing usable, no sub-processor list beyond a single mention of AWS, no statement on whether customer data trains models, and no hosting country. The privacy policy, in force since 16 November 2023, reads as a generic website policy and sits oddly beside terms refreshed on 17 April 2025 that do carry real GDPR commitments.
The free plan is genuinely free and genuinely permanent, which makes evaluation cheap, but its quotas, one application, 50 messages a month, data cleared after 30 days of inactivity, mean any real use pushes you to Starter at 29 USD per month within days.
Recommended for non-technical teams and mid-market enterprises building knowledge agents on private data. Regulated organizations can look too, but only after obtaining in writing what the site does not publish: certifications, sub-processors, hosting locations and a clear position on model training.
- Choosing a selection results in a full page refresh.
- Opens in a new window.