Zep pricing, explained

Zep prices on Episodes. An Episode is, roughly, a unit of input you feed into memory, a message or event ingested into a user or session graph, from which Zep extracts facts and maintains its temporal knowledge graph. You start with a free allotment of credits, then move onto paid plans and buy more credit as you scale. The meter ticks on the write. That makes Zep's pricing easy to start and easy to under-estimate: the cost driver is how much your agents talk, not how often you actually use what they remembered.

CognitiveX inverts that. Storing memories is free; you spend recall credits only when you read, and those are priced by depth: a Foundational recall costs 1, a Standard recall 3, a Deep recall 10. Plans are flat: Amnesiac (free), Awakened at $20/mo, and Conscious at $200/mo, with pay-as-you-go overflow and a spend cap you set. The decision rule is simple: is your agent write-heavy or read-heavy?

Zep's exact published numbers change over time, so check their site for the current credit allotments. The CognitiveX figures here are our live plans. What's structural, and won't change, is which event the meter watches: Zep watches writes, CognitiveX watches recalls.

The honest answer depends on your read/write ratio, so reason about your own traffic rather than a headline price:

• Write-heavy, read-light. A logging agent, an ingestion pipeline, a chatty support bot that records everything and queries rarely. Per-Episode metering taxes exactly the thing you do most. Here a model where storage is free and you only pay on recall tends to win.
• Read-heavy, write-light. A research assistant that holds a stable corpus and queries it constantly. Recall-metered pricing has to be sized for that read volume, so pick your plan by expected recalls, and favor Standard depth over Deep where Deep isn't earning its 10 credits.
• Balanced. Compare like-for-like: estimate Episodes/month for Zep and recalls/month (by depth) for CognitiveX, then map each onto its plan or pay-as-you-go rate. Don't forget that under CognitiveX a growing memory store costs nothing until you read it back.

A per-Episode meter and a per-recall meter both bill access to storage and retrieval. CognitiveX charges for recall because recall is where the work happens, but the memory you're recalling from isn't a passive store. It's the Large Cognition Model (LCM), where the memory is the model.

Concretely, a recall draws on a four-tier memory (semantic, episodic, procedural, foundational), with pattern detection and salience scoring deciding what surfaces, overnight dream consolidation that compresses and links memories while the system is idle, and reflection / introspection so the memory has a view of itself. An LLM does query → model → response → forget; the LCM closes the loop: query → living memory → reasoning → learning → evolution. It speaks MCP, so any agent (and any swappable LLM) recalls through it. That's why depth is a real pricing axis: a Deep recall is doing more cognition, not just pulling more rows.

This is cognitive infrastructure that makes every LLM smarter, not another LLM wrapper. Plug it in over MCP, keep your model swappable, and never get taxed for remembering.

Start building →Try iCog →

See the full plans on pricing · switching off Zep? Read the Zep alternative · or browse every comparison.