How it works
From idea to a live healthcare agent.
Six stages, each with its own tool. Build and evaluate privately, get your assurance in order, then go live only after we have reviewed your agent.
How it works
Six stages, one platform.
Each tool takes its name from bone biology. Osteoblasts lay down new bone, osteoclasts test and remodel it, and osteocytes maintain it.
- 1BuildOsteoblast
Sign up free, build from your own content, and preview your very own app on your phone.
Always free - 2EvaluateOsteoclast
Test and remodel your agent against a fixed question set, and review every answer.
Founder plan · beta - 3Assure and complyOsteocite
Get your assurance documents drafted and keep them in one place.
Founder plan · beta - 4Submit for reviewLive review
We review your agent individually before any patient can reach it.
One-off review
Then, once approved
Give it everything you would give a new colleague.
A simple first version takes about 10 minutes. After that, Build is where the quality of your agent is decided. Upload as much as you can, take your time over the passages, and talk to it until it sounds like your service.
Brief it like a new member of staff, and see exactly how the agent is instructed.
Upload leaflets, write your own text or add single questions and answers.
Set reading level, warmth and length, and write the lines it must never cross.
Name, logo and welcome message, with a live phone preview.
- Define it
- Who it is for, what it helps with and where it stops.
- Upload your content
- Patient leaflets, clinical guidelines, lectures, health promotion material and your own knowledge.
- Review the passages
- See how your content has been split up, fix the wording and approve what the agent may use.
- Add your own guardrails
- Ours are already in place. Add rules for your use case and see every one that is in force.
- Design how it looks
- Name, colours, welcome message and tone of voice.
- Try it
- Preview your very own app, free. Talk to it on a simulated phone, or scan a code and use it on your own, through the real patient pipeline.
Osti-judge gives a first opinion. Yours is the one that counts.
Every answer is scored by our judging model, then by you. Its judgement is there to help you read quickly, not to replace clinical review.
- GenerateOsti-gen drafts a set of realistic patient questions for your agent.
- Edit and lockYou review and edit the set, then fix it so every run is comparable.
- RunEach question goes through the real patient pipeline and every answer is captured.
- JudgeOsti-judge scores each answer on safety and quality.
- ReviewYou and at least one other clinician accept or overrule every score, using a shared review link.
- ReportShare an evaluation report with your stakeholders.
Choose how many questions to generate. Emergency, boundary and out-of-scope questions are always included, with a fixed safety core on top.
Safety checks
- Unsafe answers
- Missed emergency signposting
- Major clinical errors
- Missing essential information
- Incorrect source
Quality, scored 1 to 5
- Accuracy
- Clarity
- Relevance
- Completeness
- Sources
- Usefulness
Every evaluation needs at least two clinicians. Review by more than one clinician is the gold standard, and the simplest way to reduce individual bias.
Nine documents drafted from your build and evaluation.
Osteocite Assist writes the first drafts for you to review and sign off. The dashboard holds them alongside everything else your assurance process needs. Agents built on Ostify are aligned with DSPT and DTAC, and our own DCB0129 is published for you to reference.
Each record is drafted from your agent. Read it, edit anything that isn’t right, then sign it as yours.
- Intended Purpose Statement
- Device Classification Record
- Scope Boundary and Crisis Response Specification
- Source Document Register
- Clinician Review Record
- Evaluation Report
- Transferred Controls Acceptance
- Content Rights Warranty
- Change and Re-Review Record
No agent reaches a patient until we have reviewed it.
We check your content and guidelines, confirm the agent stays outside medical device territory, and try to make it misbehave, including with prompt injection. Approval is not a safety guarantee. That decision stays with your clinical safety officer.
Submit when you are ready. The readiness checklist shows what the reviewer will look at.
- Review fee
- Typically £800–£1,000A one-off fee that varies with your agent’s size and scope. We will quote before we start, or apply for Osticellerate to pay nothing upfront.
- After changes
- Re-approvalEdit your content or key settings and the agent goes back to review before patients see the change.
Pilot with real patients.
Once your agent passes review, you get a dedicated link and a generous message allowance, so you can pilot it with real patients and learn what works before you scale.
- Your link
- A dedicated patient linkShare it wherever your patients will find it.
- Your messages
- A generous message allowancePlenty of messages to run your pilot, included with approval.
- After the pilot
- Move to productionWhen you are ready to scale, move onto Ostify production.
Scale in production, and stay in control of cost.
After your pilot, your agent moves onto Ostify production. All our LLM products run on foundational models in our secure environment.
- Packages
- A set number of conversationsChoose a package with a predefined number of patient conversations each month.
- Pay-as-you-go
- Top up as you goNo commitment. Buy credit in advance and use it as your patients need it.
- Billing
- Always prepaidYou pay before you use it, so there are no surprise bills and you remain in control.
- Where it runs
- Our secure environmentFoundational models hosted in Ostify’s secure health cloud.
Start with the content you already trust.
Your agent stays private in the sandbox until you decide to take it further.