The
Parable
Memory Embedding Infrastructure for agents, copilots, and any application that currently uses embeddings for search, retrieval, or recommendation. Palo Bloom produces vectors that carry temporal position, personal context, and experiential meaning as first-class dimensions instead of metadata.
Palo Bloom is currently in development. The architecture, pricing, and API surface described here reflect what launches. Join the waitlist.
The Palo Engine Family
Choose based on the cost, latency, and depth of memory your application requires.
Palo Nano
High-frequency, low-complexity memory tasks. Consumer products, mobile applications, lightweight personalisation. Fast, low-cost, minimal compute overhead. The right choice where latency and cost dominate over recall depth.
Learn more »Palo Bloom
The default for most agent and copilot deployments. Full episodic memory with strong recall across sessions. Balanced capability and cost. Designed for the majority of production use cases where genuine memory matters.
Learn more »Palo Research
Long-horizon memory synthesis across extended personal history. Complex reasoning over time. Built for clinical and research applications, enterprise knowledge systems, and robotics memory backends where depth and precision are non-negotiable.
Learn more »
As a conversational memory layer
Most agent integrations treat memory as context management. They buffer recent messages, trim the window when it fills, and rely on the LLM to infer continuity across sessions. This is functional. It is not memory.
Palo sits between your application and your LLM, encoding each interaction as experience rather than logging it as text. When your user sends a message, Palo retrieves what is episodically relevant -- not what is semantically nearest -- and passes that context forward. Bring your own vector store, your own LLM. Palo handles encoding, storage, and episodic retrieval through a single API.
As Memory Embedding Infrastructure
Any application currently using standard embeddings for similarity search can use Palo embeddings instead. Standard embeddings capture semantic meaning at a point in time. Palo embeddings carry temporal position and personal context as part of the vector itself.
A content recommendation system using Palo embeddings can surface items relevant to where a user is in their experiential arc -- not just what they have engaged with before. A knowledge base using Palo embeddings can weight results by personal relevance and recency, not only semantic similarity.
As a recommendation backbone
Product recommendation, content ranking, personalisation engines. Any system where understanding the evolution of a user's interests over time produces better results than a static preference model.
Palo's episodic structure means the recommendation signal improves with time rather than plateauing at preference extraction. The system understands where a user has been, not only what they like.
How Palo Bloom differs
Most memory infrastructure in production today falls into two categories. Palo Bloom is neither.
| Fact-extraction systems (e.g. Mem0) | Graph-based systems (e.g. Zep) | Palo Bloom | |
|---|---|---|---|
| Memory model | Extracted facts | Temporal knowledge graph | Encoded experience |
| What is stored | Discrete propositions extracted from text | Entities and relations tracked over time | Experiential representation with temporal position |
| Retrieval | Semantic similarity | Graph traversal + semantic search | Episodic traversal grounded in personal timeline |
| Temporal awareness | Timestamps as metadata | Facts tracked as they change | Time as a first-class dimension of every memory |
| Embedding use | Retrieval only | Graph construction + retrieval | First-class episodic embeddings for any vector application |
Palo Guard
For deployments where what gets remembered matters as much as how well it is remembered.
Palo Guard is the cognitive defense layer developed by Mpalo Mind Defense. It handles manipulation-pattern detection, privacy-first memory policies, and interaction safety defaults. Palo Guard launches after Palo Bloom and is available as an add-on.
Robotics and embodied AI
Palo Research is the intended memory backend for physical systems that need to remember people, places, and tasks across extended operational periods. This capability is not available at launch. We are listing it here because it is the direction -- and we think accountability requires saying what you are building toward before you have built it.
