Private AI built for regulated industries
Astro is RocketDocs' private generative AI engine. It runs inside our environment, drafts responses using only your approved knowledge base, and never sends your data to a third-party model provider. For financial services, healthcare, life sciences, and enterprise tech teams, this is the only AI architecture that holds up under audit.
- SOC 2 Type II + ISO 27001 Certified
- Trusted since 1994
- 4.8 / 5 on G2
Why private AI matters
Why private AI matters for regulated industries
Most response management platforms have a generative AI feature. Almost none of them are private. When Loopio drafts a response, your data flows through a third-party model provider. Same with Responsive. Same with Qvidian's AI Assist. Your knowledge base, your customer information, and your historical responses leave your environment to generate every answer.
For regulated teams, that is a non-starter. Compliance review boards reject it. Procurement teams flag it. Customer security questionnaires explicitly disallow it.
RocketDocs solves this by running our own model in our own environment. No third-party LLM provider sees your data. The generative AI that drafts your responses is private by design, not by configuration toggle.
How Astro works
A three-layer approach. Generative AI is the last resort, not the first response.
Astro uses a three-layer AI approach. Each layer handles different types of questions, applied in priority order.
Exact-match autofill
Astro starts by looking for direct matches in your approved content library. Questions that have been answered before, with content that has been reviewed and approved, are filled in automatically. This is the fastest, most accurate path. No AI guessing involved.
Context-aware similarity search
When an exact match is not available, Astro uses similarity search to find the closest approved answer. This catches the cases where the question has been asked before in slightly different language. The library answers, not the model, drive the response.
Private generative AI
For questions that are genuinely new, Astro uses private generative AI built on Llama 3.1, hosted inside the RocketDocs environment. The model only references your approved knowledge base, never the public internet, never another customer's data. Every generative response is flagged for human review before it is final.
The model
Why Llama 3.1, hosted privately
Astro runs on Llama 3.1, an open-source large language model from Meta, hosted entirely inside the RocketDocs environment. This combination gives regulated customers what they actually need: a model whose architecture, training data origins, and behavior are documented in the open, deployed in a way that keeps customer data out of any third party's hands.
- Open-source model: documented architecture and training methodology, not a black-box API
- Hosted privately: the model runs inside our environment, not on Meta’s infrastructure or any external API
- No data shared with the model provider: Meta does not see your prompts, your responses, or your knowledge base
- No cross-customer training: your data is not used to train the model, full stop
Where your data lives
All customer data stays inside the RocketDocs environment
All customer data lives inside the RocketDocs environment. Encryption at rest uses AES-256. Encryption in transit uses TLS 1.2 or higher. Access is controlled through SSO, role-based permissions, and immutable audit trails. No data is ever sent to OpenAI, Anthropic, Google, or any other external model provider.
- AES-256 encryption at rest, TLS 1.2 or higher in transit
- SSO and role-based access controls
- Immutable audit trail across every Astro action
- No external model provider in the data path
Astro vs. third-party AI
Astro vs. third-party AI
Here is what private AI changes in practice.
- Data residency: Your knowledge base and responses stay inside the RocketDocs environment. Third-party AI sends data through external providers.
- Audit trail: Every Astro action is logged in your audit trail. Third-party AI logs are managed by the model provider, not by you.
- Compliance: Astro is designed to support SOC 2, ISO 27001, HIPAA-aligned workflows, and 21 CFR Part 11 expectations. Third-party AI complicates each of these reviews.
- Customer data isolation: Your data trains nothing. Astro responses are generated only from your approved content. No cross-customer model exposure.
Astro inside Microsoft Teams
Astro inside Microsoft Teams
Astro is also available as a chatbot inside Microsoft Teams. Sales reps, SMEs, and support teams can query your approved content library directly from Teams, using natural language. Astro returns answers with source citations, so the team always knows where the information came from. Permissions follow the user, so each person only sees content they are authorized to access.
- Natural-language queries from inside Microsoft Teams
- Every answer returned with source citations
- Permissions follow the user across every access point
- Same private AI engine, same audit trail, same compliance posture
Compliance posture
Built for compliance from day one
Astro is not generative AI added to a legacy platform. It is generative AI engineered into a response management platform whose customers have always required a high compliance bar. The architecture decisions, the audit trail integration, the permission model, and the data flow controls were designed together.
- SOC 2 Type II and ISO 27001 certified
- AES-256 encryption at rest, TLS 1.2 plus in transit
- Workflow patterns aligned to 21 CFR Part 11 expectations
- HIPAA-aligned content handling for healthcare deployments
- Granular permissions enforced at the content record, project, and user level
What customers say
Trusted by the teams whose responses cannot be wrong
The tool itself is very simple and direct. I've trained a lot of people on this and they're like, that's all I have to do? It's the way that RocketDocs works with Word. It's very similar to what they're used to. It's very user friendly.
RocketDocs has competitors in the space. But none of them can do what RapidDocs does. I haven't found any that are as good in product suite. So RapidDocs, from my perspective, is pretty unique. It's a great tool. It can save you time. It can help you to do things a lot easier.
Problems are the same for all RFP teams: finding the correct data at the right time, and organizing data into useful libraries and subtopics. RocketDocs allows us to manage more than 10 different lines of business and keep our data organized and structured.
After over 20 years of using different RFP database management systems, I am impressed with the usability and ease of organization in the system. The speed with which my team can locate and update responses is impressive.
Cycle time on enterprise DDQs dropped from six weeks to under two. The private-AI architecture is the only reason our security team ever signed off on adding generative AI to the response workflow at all.
We run all of our institutional questionnaire responses through RocketDocs. Multi-affiliate library structure handles our three lines of business cleanly; SME assignment and review cycles keep content accurate without anyone having to babysit it.
The Excel multi-tab handling is the feature that closed it for us. SIG Lite, SIG Core, CAIQ, our own customer questionnaires — all multi-tab, all native. The other platforms we evaluated either flattened the tabs or charged extra for the capability.
The audit trail is what finally got us off the spreadsheet-and-email pattern. When 21 CFR Part 11 reviewers ask who approved each answer and when, we have a real answer instead of digging through Slack.
FAQ
Frequently asked questions
What model does Astro use?
Astro is built on Llama 3.1, an open-source large language model from Meta, hosted privately inside the RocketDocs environment. The model runs on our infrastructure, not Meta’s, and Meta does not have access to your prompts, your responses, or your knowledge base.
Does Astro use OpenAI, Anthropic, or any other third-party AI provider?
No. Astro does not route customer data through OpenAI, Anthropic, Google, or any other external model provider. The generative AI that drafts your responses runs inside the RocketDocs environment.
Will my data be used to train the model?
No. Customer data is not used to train the Astro model. Your knowledge base, your responses, and your customer information are isolated to your environment.
How accurate is Astro?
Astro is built to prioritize accuracy over creativity. Approved content is always preferred to generated content. Similarity search is preferred to generation. Generation is the last resort and every generated response is flagged for human review. This three-layer architecture means most responses come from your reviewed library, not from the model.
What happens when Astro does not know the answer?
Astro never invents an answer. When the question cannot be matched, similarity-searched, or generated from your knowledge base, the question is left blank and routed for SME input through the standard workflow. There is no hallucination on missing information.
Is Astro available on mobile?
Yes. Astro is accessible through the RocketDocs platform, the Microsoft Teams app, and any device that supports Microsoft 365. Mobile-responsive interfaces are supported across all access points.
How does Astro handle confidential or restricted content?
Permissions enforced at the content record level apply to Astro queries. If a user does not have access to a piece of content, Astro will not return it in a response, surface it in a similarity search, or use it as context for a generative draft.
Is Llama 3.1 audit-ready for regulated industries?
Yes. Llama 3.1 is open-source with documented architecture and training methodology, which makes it materially easier to audit than closed-source commercial APIs. Combined with private hosting, granular permissions, and immutable audit trails, the deployment meets the standards regulated industries require.
Ready to see Astro in action?
A specialist will walk you through how Astro handles a real questionnaire from your industry, with the data flow, audit trail, and permission model demonstrated end to end.