Ricord vs Zep: AI Memory API Comparison (2026)

TL;DR

Zep is one of the longest-running AI memory startups and built Graphiti, a respected bi-temporal knowledge graph. They also deprecated their open-source Community Edition in 2024 — leaving self-hosted users to either migrate to Zep Cloud (paid) or find alternatives — and rebranded the product around "context engineering," which obscures the underlying memory primitive.

Ricord ships a knowledge graph on every paid tier starting at $15/mo (annual) with auto-generated wiki pages, automatic conflict resolution, a drop-in OpenAI-compatible proxy, and SDK + MCP server under MIT — so the open-source path stays open.

Quick comparison

The Community Edition deprecation

In 2024, Zep deprecated its open-source Community Edition. Teams that had self-hosted Zep CE in production were given two paths: migrate to Zep Cloud (paid) or rebuild on a different memory layer. There was no clear migration tooling.

This pattern is becoming common — open-source projects used to seed developer adoption, then deprecated once the cloud product is launched. It's a rational business move. It also means that betting infrastructure on a vendor's OSS posture carries structural risk.

Ricord's posture: the SDK and MCP server are MIT-licensed and stay that way. The hosted API is a paid service, but you can self-host the client surface area against any S3-compatible store and Postgres — the open-source path is a permanent option, not a marketing teaser.

The "context engineering" rebrand

Zep's 2025 messaging pivoted to "context engineering" — framing the product as a way to engineer better prompt context rather than a memory layer specifically. The underlying tech (Graphiti, bi-temporal graphs) didn't change.

This is a positioning choice, not a technical one. If you want a memory layer, you're looking for: durable storage, recall across sessions, conflict resolution when facts change, and a way to read what the system thinks it knows. Calling that "context engineering" doesn't add capabilities — it changes how the product is searched for.

Where Ricord wins

1. Auto-generated wiki pages

Zep gives you a search API on top of a graph. Ricord gives you the search API and a browsable wiki — every entity in the graph gets its own markdown page with backlinks, aliases, and a contradiction history. You can readwhat your agent has learned, the same way you'd read an Obsidian vault. No other memory API ships this surface.

2. Automatic conflict resolution

Graphiti's bi-temporal edges let you query "what was true at time T," which is powerful. But when a fact contradicts a stored one in the present, neither system resolves it for you. Ricord detects the contradiction, deprecates the stale fact with a timestamp, and stores the correction as canonical.

3. Sub-second recall + production reliability

Ricord runs on Cloud Run with managed autoscaling and published uptime targets. Zep's container reliability has been publicly criticized (container startup failures in Zep CE issue tracker, superficial deployment documentation). For production agents handling real user load, the reliability story matters as much as the algorithm story.

4. LLM proxy + BYOK

Ricord ships an OpenAI-compatible proxy. Change your base_url to api.ricord.ai and your existing code gets memory automatically. Bring your own model keys (Anthropic, OpenAI, Google, local). Zep has no equivalent proxy.

5. Pricing transparency

Ricord's tiers ($15/$39/$79 annual) are listed on the website. Zep's pricing requires a sales contact. For teams evaluating quickly, knowing the cost without a phone call matters.

Where Zep wins

1. Bi-temporal edges via Graphiti

Graphiti's first-class bi-temporal modeling — every edge has both validity time and ingestion time — is genuinely sophisticated. For domains where you need to query "what did we believe at time T1 about an event at time T2," Graphiti is well-suited. Most production agents don't need this, but if you do, Zep is a fair pick.

2. Longer track record

Zep has been in the memory market since 2023, with more deployment scars and more written documentation about long-term agent memory behavior. If you value "has been doing this longer," Zep wins that axis.

3. SOC 2 Type II

Zep has SOC 2 Type II. Ricord's is in progress. If your procurement requires Type II in hand today, that's a real constraint.

Who should choose what

Choose Ricord if you:

• Were burned by the Zep CE deprecation or worry about the next OSS deprecation.
• Want to read what your agent has learned (browsable wiki pages).
• Need automatic conflict resolution (not just bi-temporal storage of both facts).
• Want pricing visible on the website without a sales call.
• Need a drop-in LLM proxy with BYOK.
• Want first-class MCP support across Claude Code / Desktop / Cursor / Codex.

Choose Zep if you:

• Need Graphiti's specific bi-temporal edge model for your domain.
• Already have a Zep Cloud deployment in production.
• Need SOC 2 Type II in hand today.

Last verified May 28, 2026. Zep pricing and feature claims sourced from Zep's public marketing site and GitHub. Corrections welcome — email team@ricord.ai.