The Ontave index.
Build agents. Choose your framework.
Keep operational data grounded in evidence.
Build agents
Start with a working tool. Choose the framework around the workflow.
How to build an AI agent in TypeScript
A read-only agent tutorial with a bounded tool loop, fictional records and offline tests.
Choosing the best AI agent framework for your application
Compare AI SDK, LangGraph, Mastra, Pydantic AI and CrewAI by the work each controls.
Using LangGraph and Ontave in the same application
An architecture for execution checkpoints, durable evidence, explicit review and accepted-data reads.
Design the data layer
Make context, memory, evidence and accepted records work together.
Agent memory vs. a database: what belongs where?
Separate conversation context, execution state, retrieved passages and accepted business records.
Structured LLM output is only the start of validation
Check schema, source support, identity, permissions and current dependencies before committing facts.
Design an AI document-processing pipeline with reviewable records
Preserve document revisions, handle retries, stage extracted claims and query accepted facts.
Retrieval finds context. Ontave governs records.
Where retrieval, ordinary databases and an evidence-to-record framework each fit.
A changing model. A stable application contract.
Add an unfamiliar concept while existing applications keep their pinned interfaces.
Review and evaluate
Know what was approved, what was measured and what still needs work.
Human-in-the-loop agents: design an approval that means something
Bind review to evidence, identity, a specific proposal and the effect a person is approving.
How to evaluate an AI agent beyond a successful demo
A concrete test plan for tools, task correctness, recovery, cost and human review effort.
A review boundary for agent workflows.
Follow a solar installation message from preserved evidence to accepted operational data.
Start with the mechanism. Evaluate it against your own workflow.
Open the developer guide