Master this essential documentation concept
Bring Your Own Model - an approach that allows organizations to connect their own self-hosted AI models to a platform rather than relying on the platform's built-in third-party AI services.
Bring Your Own Model - an approach that allows organizations to connect their own self-hosted AI models to a platform rather than relying on the platform's built-in third-party AI services.
When your team integrates a self-hosted model into your existing toolchain, the setup process rarely happens in silence. Engineers walk through configuration steps on calls, security reviews get discussed in recorded meetings, and onboarding sessions cover how your specific model endpoints connect to downstream systems. That institutional knowledge lives in those recordings — until someone needs it six months later.
The challenge with video-only documentation for BYOM workflows is that model configurations change. Endpoint URLs get updated, authentication methods rotate, and the reasoning behind specific architectural decisions fades quickly when it's buried in a 90-minute onboarding recording. New team members can't search a video for "how we handle token limits" or "why we chose this inference endpoint over the default."
Converting those recordings into structured, searchable documentation means your BYOM setup — the connection logic, the security considerations, the model-specific parameters your team settled on — becomes something anyone can reference, update, and version alongside your actual configuration files. For example, when a new engineer needs to understand why your team routes certain request types to your self-hosted model instead of a platform default, they can find that decision documented directly rather than scrubbing through recordings.
If your team regularly records technical walkthroughs or architecture reviews around your AI integrations, see how you can turn those into living documentation →
A hospital system wants to use AI to help clinical staff generate and summarize patient care documentation, but HIPAA compliance prohibits sending Protected Health Information (PHI) to external cloud-based AI services like OpenAI or Anthropic.
BYOM allows the hospital to connect their internally hosted fine-tuned medical LLM (running on on-premises GPU servers) to their documentation platform, so AI-assisted writing and summarization occurs entirely within the hospital's own network perimeter.
["Deploy a HIPAA-compliant LLM (e.g., a fine-tuned LLaMA 3 model) on the hospital's internal Kubernetes cluster with no outbound internet access.", "Configure the BYOM adapter in the documentation platform (e.g., Confluence) to point to the internal model's REST API endpoint using an internal DNS hostname.", 'Set up mutual TLS authentication between the documentation platform and the self-hosted model endpoint to ensure only authorized services can invoke the model.', 'Validate data flow with network traffic analysis to confirm zero PHI egress before enabling AI features for clinical staff.']
Clinical documentation teams gain AI-assisted drafting and summarization capabilities while maintaining full HIPAA compliance, with audit logs confirming all inference requests remain within the hospital's private network.
A global investment bank's compliance team needs AI assistance to draft and review regulatory filings (MiFID II, Basel III reports), but generic cloud AI models produce inaccurate financial regulatory language and the firm cannot risk confidential trading data leaving their environment.
BYOM enables the firm to connect their proprietary compliance-tuned LLM — trained on internal regulatory documents and approved terminology — to their documentation platform, replacing generic AI suggestions with highly accurate, domain-specific outputs.
["Fine-tune a base model (e.g., Mistral 7B) on the firm's internal corpus of approved regulatory filings, compliance manuals, and legal glossaries using an air-gapped training environment.", "Deploy the fine-tuned model behind the firm's internal API gateway with role-based access control restricting usage to the compliance and legal departments.", "Register the model endpoint in the documentation platform's BYOM configuration, mapping it to AI-assist features used during document drafting and review workflows.", 'Establish a model versioning protocol so compliance officers can track which model version generated which document sections for audit trail purposes.']
Regulatory filing drafting time decreases by an estimated 40%, with compliance officers reporting significantly fewer manual corrections needed compared to outputs from generic cloud AI models, and zero incidents of confidential data exposure.
A European manufacturing conglomerate must comply with GDPR data residency requirements, meaning employee and operational data used in internal documentation cannot be processed on servers outside the EU — ruling out most US-headquartered AI cloud providers.
BYOM lets the enterprise deploy separate self-hosted model instances in EU-based data centers for each regional subsidiary, connecting each instance to the documentation platform so that AI assistance for each team is processed exclusively within the legally required geographic boundary.
['Provision GPU-enabled virtual machines in EU-West and EU-Central Azure regions (or equivalent on-premises hardware) and deploy isolated model instances (e.g., Ollama serving Mistral) on each.', 'Configure geolocation-aware BYOM endpoint routing in the documentation platform so that users authenticated from EU tenants are directed to the EU-resident model endpoints.', 'Implement data residency attestation logging at the model API layer to generate compliance reports showing that all inference requests originated and were processed within EU boundaries.', 'Conduct a GDPR Data Protection Impact Assessment (DPIA) documenting the BYOM architecture as evidence of technical measures for data residency compliance.']
The enterprise passes GDPR audit reviews with documented proof of EU data residency for all AI-assisted documentation workflows, avoiding potential fines and enabling AI productivity tools to be rolled out to EU employees who were previously excluded.
An aerospace defense contractor's engineering teams need AI assistance to write and review classified technical specifications and system design documents, but classified networks are physically air-gapped from the internet, making any cloud-based AI service completely inaccessible.
BYOM enables the contractor to run a self-hosted model on the classified network itself and connect it to an on-premises instance of their documentation platform, bringing AI capabilities into the air-gapped environment without any network bridge to external services.
["Obtain and transfer approved open-weight model weights (e.g., LLaMA 3 via approved media transfer process) onto the classified network's local model registry following security protocols.", 'Deploy the model using an on-premises inference server (e.g., vLLM or llama.cpp) on hardware certified for the classification level of the network.', 'Install and configure an on-premises instance of the documentation platform (e.g., Confluence Data Center) and register the local model endpoint via the BYOM configuration interface.', 'Conduct security certification testing to verify no covert channel exists between the AI model inference layer and any unclassified network segment before granting user access.']
Engineering teams on classified programs gain AI-assisted documentation capabilities for the first time, reducing technical specification drafting cycles while maintaining full compliance with government security accreditation requirements for the classified environment.
Most documentation platforms that support BYOM expect an API contract similar to the OpenAI Chat Completions API format. Wrapping your self-hosted model (whether LLaMA, Mistral, or a custom model) behind a compatibility layer such as LiteLLM, Ollama, or a custom FastAPI proxy ensures the platform can communicate with your model without bespoke integration code. This also makes it straightforward to swap underlying models without reconfiguring the platform.
The primary security motivation for BYOM is keeping sensitive data within your control, but this benefit is nullified if the connection between the documentation platform and your model endpoint is not properly secured. An unauthenticated or weakly authenticated model endpoint could be invoked by unauthorized parties, exposing your model's capabilities or leaking prompt data. Mutual TLS (mTLS) combined with short-lived bearer tokens provides defense-in-depth for this critical data path.
When AI-generated content appears in official documentation, organizations need to know which model version produced it, when, and based on what prompt. Logging at the BYOM gateway layer (not just application logs) captures the full inference request lifecycle, which is essential for compliance audits, debugging unexpected model outputs, and tracking model performance over time. Ensure logs are stored in your SIEM or log management platform with appropriate retention policies.
AI model behavior changes between versions — a model update that improves general quality may alter tone, terminology, or formatting in ways that affect documentation consistency. Without explicit model versioning tied to your BYOM configuration, it becomes impossible to reproduce a specific AI output or understand why documentation quality changed after a model update. Treat model versions as first-class artifacts in your change management process.
Unlike cloud AI APIs that scale elastically, your self-hosted BYOM model has fixed hardware capacity. If the documentation platform's AI features trigger simultaneous inference requests from many users — such as during a team sprint where everyone is writing docs at once — an undersized model server will experience severe latency or request queuing that degrades the user experience and may cause platform-side timeouts. Proactive capacity planning based on measured usage patterns prevents this.
Join thousands of teams creating outstanding documentation
Start Free Trial