AgentPrizm
by VUGA Enterprises LLC
The memory of record for AI agents — recall it, prove it, delete it
AgentPrizm is a hosted memory and skills API for AI agents, reachable over REST or a remote MCP endpoint. It stores confidence-scored facts with validity windows, resolves contradictions so newer facts supersede older ones, and returns an audit receipt explaining why each memory was recalled. It is aimed at teams whose support, sales or coding agents must remember context across sessions and prove what they remembered.
AgentPrizm is a hosted memory-of-record service for AI agents, launched publicly on 9 July 2026 by VUGA Enterprises LLC of Miami, Florida, founded by CEO Gene Avakyan with co-founder Victoria Unikel. It has two halves. AgentMemory stores atomic memories of six types — fact, lesson, directive, preference, contact and bookmark — each carrying a calibrated confidence score between 0 and 1, a source, tags and an optional validity window expressed as validFrom and validUntil. Recall is hybrid, combining semantic vector similarity with keyword matching, then filtered by validity window and passed through contradiction resolution in which newer facts automatically supersede older ones and expired memories are excluded unless explicitly requested. Every recall returns an audit receipt showing why each memory ranked where it did, and containers partition memory per user, organisation, agent, session or environment so production and staging never bleed into each other. GDPR-aligned right-to-forget is a single API call with verifiable deletion. AgentSkills is the second half: a governed marketplace of reusable agent procedures published as versioned SKILL.md files, discoverable by intent, forkable, with lineage preserved and secret-blocking on publish. Access is through a REST API at /api/v1/ and a remote MCP endpoint at /api/mcp authenticated with ap_-prefixed bearer keys, giving zero-install setup for Claude Code, Cursor, Claude Desktop and any MCP-capable agent, plus distribution as an OpenClaw skill on ClawHub. Migration tooling exists for teams moving off Mem0, Zep, Letta or a Postgres and pgvector stack. The service is hosted only, with no self-hosting option; the company has said it plans a source-available version of the memory engine.
A small engineering team shipping MCP-based agents that already lose context between sessions, and that needs recall it can explain and delete on request rather than an opaque vector store.
Agents keep verifiable, time-bounded memory across sessions, and every recall comes back with a receipt showing what was remembered and why.
At a Glance
- Category
- Agent Development
- Pricing
- Freemium, Subscription, Usage-based
- Target Market
- Enterprise Developers, AI Engineers, Startup CTOs, Platform Engineers
- Deployment
- Cloud-only, API-based
- Founded
- 2026
- Headquarters
- Miami, Florida, United States
Key Features
- ✓Confidence-weighted memory
Every stored fact carries a calibrated 0-to-1 confidence score, so an agent can weight or discard weak recollections instead of treating all context equally.
- ✓Fact-validity windows
validFrom and validUntil bound when a fact is true, so expired information drops out of recall rather than being asserted forever.
- ✓Contradiction resolution
When a newer fact conflicts with an older one, the newer supersedes it automatically, which is what stops an agent contradicting its own earlier answers.
- ✓Audit receipts on recall
Each recall returns evidence of why particular memories ranked where they did, turning retrieval from opaque context stuffing into something reviewable.
- ✓Container isolation
Memory is partitioned per user, organisation, agent, session or environment, so tenants and staging traffic cannot read each other's context.
- ✓Right-to-forget deletion
A single API call performs GDPR-aligned deletion with verifiable removal, which is the control a privacy review actually asks about.
- ✓AgentSkills marketplace
Reusable procedures publish as versioned SKILL.md files with intent-based discovery, forking, preserved lineage and secret-blocking before publication.
- ✓MCP-native and REST access
A remote MCP endpoint gives zero-install setup in Claude Code, Cursor and Claude Desktop, while REST covers any other language over HTTP.
Capabilities
Use Cases
- •Persistent memory for a coding agent
Preserve architectural decisions and conventions across sessions so the agent stops relitigating choices the team already made.
- •Support agent with cited customer history
Recall a customer's prior issues with source attribution, so answers reference real history instead of a summarised prompt blob.
- •Proving a deletion request was honoured
Delete a customer's stored facts through one API call and produce the verifiable receipt a compliance reviewer asks for.
- •Standardising team procedures as skills
Publish versioned SKILL.md workflows to a private registry so every agent runs the same reviewed procedure rather than a copied prompt.
- •Migrating off a homegrown vector store
Move existing memories from Mem0, Zep, Letta or Postgres with pgvector onto a managed layer with governance primitives built in.
Ideal For
Best For
- ✓Giving a coding agent persistent recall of architectural decisions across separate sessions and repositories
- ✓Letting a support agent recall a customer's history with source citations rather than restating context in every prompt
- ✓Handling facts that expire, using validity windows so an agent stops asserting last quarter's pricing
- ✓Satisfying GDPR right-to-forget on agent memory with a single delete call and verifiable removal
- ✓Sharing reusable agent procedures across a team as versioned, forkable SKILL.md files
Not Ideal For
- ✗Regulated healthcare workloads — the service is explicitly not for protected health information, and HIPAA certification is a roadmap item rather than a current attestation
- ✗Enterprises whose security review requires SOC 2 Type II today; the company lists SOC 2 Type II only on the roadmap at the Scale tier and above
- ✗Teams that must self-host their memory layer, since AgentPrizm is a hosted service with no self-hosting option and the promised source-available engine is not yet shipped
- ✗Risk-averse buyers who need production references or third-party validation, because the only named production users are the vendor's own portfolio companies and there is no independent review coverage
Deployment
Market Analysis
Pros
- ✓Governance features that most agent-memory layers lack — validity windows, contradiction resolution and an audit receipt explaining every recall
- ✓Transparent published pricing including a genuinely usable free tier and explicit per-1,000 overage rates, so cost can be modelled before signing anything
- ✓Zero-install MCP integration for Claude Code, Cursor and Claude Desktop, plus a plain REST API, keeps adoption cost low for a small team
- ✓Documented migration path from Mem0, Zep, Letta and Postgres with pgvector, which lowers the switching cost from an existing stack
- ✓Container partitioning per user, organisation, agent, session and environment is a sensible multi-tenant default rather than an enterprise upsell
Cons
- ✗No SOC 2 and no HIPAA certification, and protected health information is explicitly prohibited — that rules out a large share of enterprise buyers today
- ✗Hosted only, with no self-hosting option; the source-available memory engine the company said it would release remains a promise rather than shipped code
- ✗There is effectively no independent evidence about this product: zero Hacker News mentions, no G2, Capterra, TrustRadius or Product Hunt review presence found as of September 2026, and the only coverage is the company's own syndicated launch release
- ✗The only production references disclosed are the founder's own portfolio companies — LogisticsPrizm, ViralPrizm and VUGA TV — not arm's-length customers
- ✗No disclosed funding and an undisclosed team size behind a service that would sit on the critical path of an agent's memory, which is a real continuity risk to weigh
- ✗Performance claims such as confidence calibration and contradiction resolution quality are unbenchmarked by any third party
Pricing
Hobby
$0
- ✓10,000 memories
- ✓4,500 recalls per month (hard cap)
- ✓7-day audit log
- ✓2 agents, 1 environment
- ✓10 private and 10 public skills
Starter
From $19/mo
- ✓100,000 memories
- ✓15,000 recalls per month included
- ✓90-day audit log
- ✓5 agents, 3 environments
- ✓Unlimited skills
- ✓14-day free trial
Builder
From $79/mo
- ✓400,000 memories
- ✓65,000 recalls per month included
- ✓1-year audit log
- ✓25 agents
- ✓Priority support
Scale
From $249/mo plus usage
- ✓1,000,000 memories
- ✓225,000 recalls per month included
- ✓Unlimited agents
- ✓Audit export
- ✓SOC 2 Type II on roadmap
Enterprise
Contact for pricing
- ✓Custom memory and recall limits
- ✓Custom DPAs, residency and retention
- ✓SSO, SCIM and audit export
- ✓Solutions engineer and 24/7 support
- ✓Volume pricing
List pricing is published, which is unusual in this category: a free Hobby tier with 10,000 memories and a hard cap of 4,500 recalls a month, then $19, $79 and $249 a month for Starter, Builder and Scale, each with an included recall allowance and a 14-day trial. Paid plans meter overage at $1.10 per 1,000 recalls and $0.50 per 1,000 ingests, so cost scales with agent chattiness rather than seats. Enterprise adds SSO, SCIM, custom residency and retention at a negotiated price. Annual billing is advertised at a lower effective monthly rate.
Security & Compliance
Sources
This page was written from 5 sources, 2 on domains other than agentprizm.com.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
OpenAI Agents API
The Codex agent harness behind one API call — managed sessions, sandboxes and subagents
Letta
Open-source platform for stateful AI agents with persistent memory, now with a TypeScript Agents SDK for embedding them in your own apps
Kiro Crew
Open-source workspace for coding agents that keep working after you close the laptop
Naïve
Autonomous company infrastructure — compute, payments, identity and incorporation for AI agents behind one API