Without AgentPrizm
Pull the renewal numbers for Arc.
Sure — what's Arc? Could you share the account ID?
Schedule a call with Rachel.
Who's Rachel? What email?
…I told you yesterday.
The AI agent memory layer
Trusted memory
Trusted memory
for AI agents.
AgentPrizm gives your AI agents persistent, governed memory — with confidence scores, fact-validity windows, audit receipts, and right-to-forget. Drop in one API. Your agents stop starting from zero.
REST API + MCP
Audit receipt on every recall
Fact-validity windows
Right-to-forget
GDPR-aligned
Battle-tested in our own products before we ask you to trust it —
LogisticsPrizm
ViralPrizm
VUGA TV
We run AgentPrizm across our own portfolio — from freight dispatch (LogisticsPrizm) to content and media ops (ViralPrizm, VUGA TV). It's the memory layer we built because our own agents needed one.
The problem
Stateless agents are expensive amnesiacs.
Every conversation starts from zero. Users repeat themselves. Facts go stale. Agents contradict last week's answer this week. RAG plus a vector DB can retrieve documents — but it doesn't know what's true now, what expired, what superseded what, or why a memory was used.
Repeated context
Users say the same thing again and again — and the agent still asks.
Stale facts
Agents use old context as if it were still true, long after it changed.
No proof
Nobody can explain why a memory showed up in the prompt.
What changes
.jpg)
What changes when your agent remembers?
With AgentPrizm
Pull the renewal numbers for Arc.
Arc renewal Q3 — blocked on procurement last cycle. Rachel (now VP Eng, owns budget) prefers async. Slack thread or a draft email?
Schedule a call with Rachel.
She's async-only this quarter — confirmed across 4 prior threads. Want a draft agenda doc instead?
Three steps to ship
.jpg)
Wire trusted memory into your agent in an afternoon.
Call it over HTTP, or connect via MCP. Extraction, deduplication, contradiction handling, and decay happen for you — no taxonomy to maintain.
1
Ingest a turn
Send a transcript, tool result, CRM event, ticket, or doc. AgentPrizm extracts facts, lessons, and preferences automatically.
POST /api/v1/agent/conversations
Authorization: Bearer ap_...
{ "container": "user:rachel" }
2
Recall what matters
Query in natural language. Hybrid semantic + keyword recall returns confidence-ranked, validity-aware memories within a token budget.
POST /api/v1/agent/recall
{ "query": "...",
"searchMode": "hybrid",
"minConfidence": 0.6 }
3
Drop into your prompt
Get back a token-budgeted context block with a receipt for every recall. Your agent stops starting from zero.
POST /api/v1/agent/context
{ "budget": 1200 }
# → prepend to your prompt
Primitives
Memory, not just retrieval.
Storing and searching embeddings isn't enough. Agent memory needs tense, trust, and lineage. Six primitives, built because store-then-search wasn't enough — powered by our patent-pending memory engine.
confidence ≥ 0.6
Confidence-weighted facts
Every memory carries a calibrated 0–1 confidence. Below threshold it asks; above, it commits — with a receipt.
valid_from / valid_to
Fact-validity windows
"Rachel was a Senior PM" is true until March 14. AgentPrizm tracks when facts begin, change, and expire.
audit trail per recall
Auditable receipts
Every recall ships with which memories matched, why, what was filtered, and the supersede chain.
container: user:<id>
Container scopes
Per-user, per-org, per-agent, per-session. Isolation by default; cross-container queries are explicit and logged.
POST /forget
Right-to-forget
GDPR Article 17 in one POST. Soft or hard delete — audited and irreversible. The forgetting itself is recorded.
searchMode: hybrid
Hybrid recall
Semantic + keyword retrieval with query expansion and optional rerank — tuned for small, high-value memory sets.
Drop-in API
One memory layer. Two ways in.
Ingest a conversation, recall what matters, build context within a token budget. Everything else — extraction, dedup, contradiction handling, decay — happens for you.
MCP-native — zero install
Paste one config block into Claude Code, Cursor, or Claude Desktop — and your agent has memory.
{
"mcpServers": {
"agentprizm-memory": {
"type": "http",
"url": "https://agentprizm.com/api/mcp",
"headers": { "Authorization": "Bearer ap_..." }
}
}
}
REST API — any language
Call it from any stack over HTTP. One integration, audited by default.
curl https://agentprizm.com/api/v1/agent/recall \
-H "Authorization: Bearer ap_..." \
-H "Content-Type: application/json" \
-d '{"query":"renewal blockers",
"searchMode":"hybrid"}'
Where AgentPrizm wins
Built for agents in high-stakes work.
Generic memory is fine for chatbots. For agents that close revenue, resolve tickets, or defend cases — you need governance, validity, and proof.
Sales agents
Persistent account memory across reps and renewals. Your SDR-bot knows last quarter's blockers, who got promoted, and the objection you've heard four times.
Support agents
Tier-1 bots that remember the customer, the device serial, last week's workaround, and the policy that just changed. Validity windows mean no stale advice.
Legal & compliance
Case agents that cite precedent with provenance. Right-to-forget is one API call. Every recall is reproducible from its trace ID.
Coding agents
Project memory for coding agents: architecture decisions, repo conventions, prior bugs, and constraints — across sessions.
By the numbers
Honest where it matters.
6
Memory types — fact, lesson, directive, preference, contact, bookmark. Painfully small on purpose.
1
API plus an MCP server. One integration, two ways in.
0
Lines of taxonomy you maintain. We extract, dedupe, and file.
∞
Every recall feature on every plan — including the free tier.
How it compares
Retrieval gets you matches. Memory gets you trust.
A vector database stores and searches. A DIY stack means you maintain the taxonomy, the decay, and the audit log yourself. AgentPrizm ships all of it.
| Capability | AgentPrizm | Vector DB + RAG | DIY in-house |
|---|---|---|---|
| Extracts memory from raw turns | ✓ Built in | No | Build it |
| Confidence-weighted memories | ✓ Built in | Search score only | Hard to calibrate |
| Fact-validity windows | ✓ Built in | No | Build it |
| Auditable receipts per recall | ✓ Built in | No | Expensive to maintain |
| Contradiction handling & decay | ✓ Built in | No | Build it |
| Right-to-forget, audited | ✓ One API call | Manual | Risky edge cases |
| Hybrid semantic + keyword recall | ✓ Built in | 1 channel | Build it |
Trust & governance
Governed memory by default.
Audit trail, container isolation, and right-to-forget on every plan — including the free tier. We're precise about what's true today.
Audit trails
Every recall has a trace — which memories matched, why, and what was filtered.
Container isolation
Per-user, per-org, per-agent, per-session scopes. Cross-container queries are explicit and logged.
Right-to-forget
GDPR Article 17 via one POST /forget — soft or hard delete, audited.
GDPR-aligned
Signed DPA and a published sub-processor list. We do not currently hold SOC 2 or HIPAA, and the service must not be used for PHI. SOC 2 is on the roadmap.
Pricing
Free until you ship. Honest after.
No surprise upgrades. Every recall feature — hybrid retrieval, confidence, validity windows, audit trails — is on every plan, including the free tier. And the free tier is deliberately generous: 1,000 memories and 4,500 recalls every month, no credit card.
Monthly
Annual — save 17%
Hobby
$0 / forever
For prototypes and side projects. No credit card.
- 1,000 memories, 4,500 recalls / mo
- 2 agents / API keys
- API + MCP access included
- All recall features
Most popular
Builder
$24 / mo
For solo devs and indie agents with paying users.
- 250K memories
- 25K recalls / mo
- 10 agents / API keys
- Email support · 14-day trial
Scale
$207 / mo
For teams running production agents at scale.
- 1M memories
- 225K recalls / mo
- Up to 999 agents / API keys
- Priority support · 14-day trial
Enterprise
Custom
Custom retention & limits on our hosted infra.
- Custom retention & limits
- SSO / SCIM on request
- Audit export
- Talk to us about your needs
FAQ
Questions, answered.
A vector DB does retrieval — it stores embeddings and returns nearest matches. AgentPrizm adds the layer above: it extracts memories from raw turns, tracks confidence and validity over time, resolves contradictions, runs hybrid semantic + keyword recall, and ships an audit receipt with every recall. Retrieval is one part of what we do.
A REST API at /api/v1/agent/* plus an MCP server — call it from any language over HTTP, or connect via MCP. Six memory types and no taxonomy to maintain. If you run an MCP client like Claude Code, Cursor, or Claude Desktop, paste one config block for zero-install memory.
Every memory carries a valid_from and an optional valid_to. When a newer fact supersedes an older one, the old memory is marked stale and the supersede chain is recorded — so a fact that was true last quarter doesn't poison a decision this quarter.
We're GDPR-aligned: a signed DPA, a published sub-processor list, and right-to-forget via one POST /forget. We do not currently hold SOC 2 or HIPAA, and the service must not be used for PHI. See our security and DPA pages for the full picture.
Not today — AgentPrizm is a hosted service. Enterprise customers get custom retention and limits on the same hosted infrastructure. If you have specific data-residency needs, contact us.
Free until you ship, with a generous Hobby tier and no credit card. Builder and Scale come with a 14-day free trial. Critically, every recall feature is on every plan — there's no graph-tier upcharge, and API + MCP access is included on the free tier.
Give your agents a memory worth keeping.
One API — the memory of record for your agents. Confidence scores, validity windows, and audit receipts included. Free until your agents ship.