OpenAI builds and deploys advanced AI models like GPT-4o for autonomous agents and workflows.
HUPI
FreeStarting price
0Popularity
About HUPI
HUPI provides durable, provider-independent memory for AI products by acting as a drop-in gateway between existing clients and LLM vendors. It intercepts requests from OpenAI-API-compatible clients, enriches them with relevant memory retrieved from a self-hosted Postgres database, forwards the enriched request to the configured LLM vendor, and then durably stores the exchange for future use. The system is designed to avoid vendor lock-in by storing memory as plain decrypted text and metadata rather than proprietary formats. Every memory assertion is independently verified against its source before being trusted, and all retrieval decisions are recorded for auditability. HUPI supports integration with any OpenAI-compatible client without modifying application code beyond changing the base URL. It also offers a VS Code extension that captures conversations, inline edits, and code completions, routing them through the same memory infrastructure used by other tools.
Key features
- OpenAI-API-compatible gateway
- Self-hosted Postgres backend
- Memory consolidation across sessions
- Fact-checking against source text
- Audit logging for retrieval decisions
- VS Code extension with chat sidebar
- Inline code edits with diff preview
- Multi-file edit with selective acceptance
Use cases
- Persistent user memory in AI chat products
- Codebase-aware AI assistants in VS Code
- Audit-compliant AI interactions with traceable history
Pros
- Provider-independent memory storage
- Self-hosted on user's own Postgres
- Fact-checked memory assertions
- Audit trail for all retrieval decisions
- VS Code extension for code-related interactions
Cons
- Requires self-hosting infrastructure
- No free tier or hosted option mentioned
- VS Code extension limited to codebase interactions
Frequently asked questions about HUPI
What is HUPI and what problem does it solve?
HUPI is a self-hosted gateway that provides durable, provider-independent memory for AI products by intercepting and enriching requests from OpenAI-compatible clients with relevant memory stored in a self-hosted Postgres database. It solves issues like memory resets between sessions, vendor lock-in, and the complexity of building custom memory or RAG systems.
How does HUPI integrate with existing AI products?
HUPI acts as a drop-in replacement by changing the client's base URL to point to the HUPI gateway instead of the LLM vendor. No changes to application code or prompts are required beyond this one-line configuration update.
Can HUPI work with any LLM vendor?
Yes, HUPI supports integration with any OpenAI-compatible LLM vendor. Users configure their preferred vendor in a providers.yaml file, and HUPI forwards requests accordingly while storing memory independently of the vendor.
How does HUPI ensure the integrity of stored memory?
HUPI independently verifies every memory assertion against its source text before trusting it, avoiding confident but incorrect claims. It also records all retrieval decisions for auditability, allowing users to trace what was remembered and from where.
Does HUPI offer a VS Code extension?
Yes, HUPI provides a VS Code extension that captures conversations, inline edits, and code completions, routing them through the same memory infrastructure used by other clients. It integrates seamlessly with the editor's native chat and completion features.
Is HUPI open-source and self-hosted?
Yes, HUPI is open-source and designed for self-hosting. It supports deployment via Docker, Kubernetes, and Helm, and requires users to bring their own Postgres database for storing memory.
