Open by Standard
Most ERP vendors are racing to ship AI built on a single model from a single provider. That looks fast, but it locks customers into a stack that is hard to change. AI capability moves too quickly — and too unevenly — to hand control to one vendor.
Predictiv AI is provider-agnostic by design. The same chat experience runs on OpenAI, Anthropic, AWS Bedrock, Azure, Google, or a model you host yourself. Switching is a configuration change. Your prompts, embeddings, and tool definitions stay where they are.
We host the orchestration layer in your tenant. No chat content, no embeddings, no audit trail leaves your environment unless you choose to use a hosted model — and even then, the data goes only to the provider you nominate.
Challenges We Solve
Common open by standard challenges—and how Predictiv addresses them.
Capability differs by model
Document extraction is not summarisation is not long-context reasoning. The model that wins one job often loses another. The leaderboard moves every quarter.
Route the right model to the right job. Predictiv AI lets you choose model per workflow — no whole-platform commitment.
Frontier capability moves quickly
New frontier models ship every few weeks. Locked-in customers wait for their vendor to integrate them.
Switch when something better arrives. The chat experience and the tool layer don't change; only the model behind them does.
Costs vary by 10x or more
A request to a frontier reasoning model can cost ten times what a small extraction model costs. Treating all AI calls the same hides real money.
Match capability to spend. Use a fast, cheap model for routine extraction; a frontier model for one-off analysis. Predictiv AI is built for this kind of routing.
Data leaves your tenant only when you choose
Most ERP AI assumes your data flows to the vendor's chosen LLM provider. That isn't always acceptable to security, compliance, or sovereignty teams.
The orchestration layer runs in your tenant. Chat content, embeddings, audit trail, and prompts all stay in your environment. Outbound model calls are explicit and configurable.
Three reasons to stay open
Provider-agnostic model layer
Choose your provider per chat flow, per workflow, or per organisation. Today's supported providers include OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, Google Vertex, and self-hosted models.
- Per-workflow provider selection
- Configuration change, not code change
- No data migration on provider switch
- Self-hosted models supported
Open-standard tool layer
Predictiv exposes its data and processes through the Model Context Protocol (MCP) — an emerging open standard for AI tool access. Any MCP-compatible AI client can drive Predictiv.
- Open standard, not a proprietary protocol
- Documented tool catalogue
- Same role and audit enforcement regardless of which client connects
- Future-compatible — open standards survive provider churn
Tenant-resident orchestration
The orchestration layer runs in your tenant. Chat sessions, embeddings, prompts, and audit trails do not leave your environment by default.
- Self-hosted on your infrastructure
- Embeddings live on your existing PostgreSQL
- Audit trail stored in your ERP database
- Outbound model calls are explicit and configurable
Versioned prompts
Every prompt the assistant uses lives in a registry. Business owners edit them; changes are versioned, audited, role-gated.
- Edit-and-publish workflow without code changes
- Full version history
- Role-based access control on who can edit which prompt
- Test-before-publish pattern
Comprehensive Capabilities
Full-featured open by standard tools built into the platform.
Configuration, not customisation
- Provider selection per chat flow
- Model parameter overrides per workflow
- A/B test new models against current production traffic
- Cost telemetry per provider, per workflow
Standards-aligned
- Model Context Protocol for tool access
- JSON-RPC 2.0 wire protocol
- Standard JWT authentication
- Open-format embeddings on PostgreSQL
Migration friendly
- No proprietary embedding format
- Prompts are portable text, not platform metadata
- Tool definitions follow the open standard, not a vendor's spec
- Audit trail is plain database tables you can export
Use Cases
How businesses use Open by Standard to drive results.
Multi-provider sourcing
A regional bank uses one provider for compliance reasoning (long-context), another for general chat, and a self-hosted model for sensitive customer data. All three serve the same workspace. Provider selection happens at chat-flow level, invisible to users.
Provider migration
A customer who started on one provider's API tier moves to another for cost reasons. The migration is a configuration update — chat flows, prompts, and tool definitions remain unchanged. Total downtime: minutes.
Tenant-resident operation
A regulated entity requires that no operational data leave its environment. Predictiv AI runs entirely in their tenant — embeddings on their PostgreSQL, orchestration on their infrastructure, audit trail in their ERP database. The only outbound traffic is explicit hosted-model calls, configurable per workflow.
Continuous benchmarking
A finance team A/B tests a frontier reasoning model against their current production model on the same workflow. Cost telemetry shows a 6x cost difference; quality telemetry shows a measurable accuracy lift on hard cases. The decision is data-driven, not vendor-driven.
Why Predictiv for Open by Standard
True BYO across the stack
Several ERP vendors offer "bring your own LLM"; few offer it across the orchestration layer too. Predictiv lets you swap the model, swap the orchestrator (it runs in your tenant on standards), and swap the tool client. Each layer is independently replaceable.
Built around an open standard
The tool layer uses the Model Context Protocol — an emerging open standard. That means any MCP-compatible AI client can drive Predictiv with the same role and audit enforcement. Open standards survive vendor churn; proprietary protocols don't.
Configuration-grade portability
Switching providers is not a project; it's a configuration change. The same chat flows, the same prompts, the same tools — different model behind them. We've built the platform around the assumption that the model layer will keep moving, and your ERP shouldn't have to follow.
Costs you can route
Different jobs need different capability tiers. Predictiv lets you route the right model to the right job — frontier reasoning for one workflow, cheap extraction for another. Your AI spend matches your business value.
Is This Right for You?
Predictiv Open by Standard works best for organisations with these characteristics.
Company profile
- Multi-cloud or hybrid-cloud strategy
- Existing AI tooling investments across multiple providers
- Compliance, data sovereignty, or residency requirements
- Cost-conscious AI strategy with clear unit-economics expectations
Pain points
- Lock-in concern with single-provider AI products
- Need for tenant residency on data and operations
- Pressure to optimise AI spend across capability tiers
- Existing prompt libraries hard to maintain in vendor-specific formats
Scale
- Active AI experimentation with multiple models in evaluation
- Multiple use cases requiring different model capability tiers
- Roadmap to multiple model providers within the next 24 months
- Existing prompt-engineering culture inside the business
Ready to Improve Your Open by Standard?
See how Predictiv can transform your open by standard capabilities. Book a personalised demo.