Skip to content

Palo Bloom

Palo Bloom is experience-aware memory infrastructure for applications that need continuity across interaction, identity, and time. It does not generate conversational responses. Its role is to enrich input with memory-aware context before that input is passed to an external LLM, or to produce embeddings and memory representations for downstream retrieval, context assembly, and personalization workflows.

Palo Bloom operates on a bring-your-own-vector-store model, with optional Mpalo-managed storage where applicable. You bring your storage provider or use Mpalo storage, while Mpalo handles encoding, retrieval, memory traversal, and personalization through a single API call. Storage costs are either paid directly to the provider or included in the subscription, depending on the deployment.

Palo is not an LLM provider. External LLM calls and costs remain separate and use your own provider keys.

Core Concepts

Memory Encoding is Palo's primary value event. Every input is transformed into a durable, experience-aware memory representation: a latent memory embedding plus rich memory content designed for recall, rendering, traversal, and long-term adaptation.

Palo Bloom is built to support:

  • -- Memory-aware retrieval
  • -- Personalization across sessions
  • -- Temporal continuity
  • -- Memory traversal for deeper context when needed
  • -- Memory mapping for more advanced retrieval paths

Each user can accumulate memory in a way that reflects their own patterns over time, so the system becomes more aligned with that user's history, style, and usage context.

Retrieval Modes

Palo Bloom exposes three tiers of memory operation:

Memory Recall

The default retrieval path -- semantic relevance and context retrieval. Best for most production use cases.

Memory Traversal

Multi-hop retrieval across a user's personal timeline. Activated when broader historical context is needed.

Memory Mapping

Advanced multi-path retrieval for high-fidelity synthesis over extended histories. Highest depth, highest precision.

Recall is the normal production default. Traversal and Mapping are optional depth features and should be enabled selectively when the task benefits from broader temporal context.

Memory Behavior

Palo Bloom supports controlled persistence and controlled forgetting. Users can remove memory explicitly, and the system can surface older material for cleanup when it is no longer useful. This keeps the memory layer aligned with user intent while avoiding unnecessary accumulation.

What Palo Bloom Is For

Palo Bloom is meant for products that need memory to be a native part of the experience rather than an add-on. It is useful when continuity, personalization, and long-term context matter more than a single isolated response.

Palo Bloom can also power any system that requires personalization beyond what standard similarity search provides -- music recommendation, content curation, behavioral modeling. Standard similarity doesn't capture the personal connections between users and their history. Palo Bloom does.

Quickstart → Memory Engines