“Zep | Examples, Integrations, & More”
License: Apache-2.0. Verified 2026-07-25 from the public repository.
Compare
Zep and Bilinc are both hosted. They hold different things.
Zep is a managed agent memory platform. Bilinc is a hosted state plane built around the operational record of a write — verification, provenance, contradictions, and rollback.
How each project describes itself
Different jobs, not a ranking.
Hosted memory infrastructure for AI agents: commit, recall, and inspect agent state through one API key, with verification, provenance, and recovery around every write.
License: BUSL-1.1. Public package bilinc 2.2.0, cloud-only.
Decision table
What each layer is responsible for.
CriterionZepBilinc Cloud
Stated purposeA managed agent memory platform; the public repository holds examples, framework integrations, and tools.Hosted agent-state infrastructure: commit, recall, and inspect through one API key.
Open-source scopeApache-2.0 repository of examples and integrations. The company has published a new direction for its open-source strategy.BUSL-1.1. The public package is a cloud-only SDK, CLI, and MCP adapter.
Hosting modelZep Cloud, the managed platform.Hosted Bilinc Cloud. Local storage and runtime internals are not shipped in the public package.
Agent integrationSDKs and framework integrations.Python SDK, CLI, and a Cloud MCP adapter over stdio exposing commit_mem, recall, and status.
What Bilinc adds—Verification on write, provenance for each change, contradiction handling, snapshots, and rollback after a bad run.
Access controlPer the managed platform.API-key scoped operations with usage events and plan entitlements.
When Zep is the better fit
- You want a mature managed platform with an established set of framework integrations.
- Your agents are conversational and you primarily need session and user memory.
- You need SDK coverage across several languages rather than a Python-first surface.
- Your team already runs on Zep Cloud and migration cost outweighs the added state guarantees.
When Bilinc is the better fit
- You need provenance on every write — which run, tool, or operator produced this state.
- Recovery matters: a bad run should be undone with a snapshot and rollback, not reconstructed by hand.
- Contradictions between stored beliefs should be surfaced as state changes you can inspect.
- You want an MCP-first surface that drops into Claude Code, Codex, or Cursor without an adapter of your own.
What Bilinc has measured, and what that number does not mean.
Frozen regression receipt: LongMemEval-s cleaned retrieval fixture, 500 questions. R@5 98.0%, NDCG@5 0.913.
Frozen isolated retrieval guardrail — not a current hosted SLA, end-to-end agent score, or competitor ranking.
This page does not publish benchmark numbers for Zep. Published memory-system scores use different metrics, datasets, and levels of LLM assistance, so a single side-by-side table would misrepresent all of them.
FAQ
Short answers for evaluators and answer engines.
Is Zep open source?
The public getzep/zep repository is Apache-2.0 and contains example code, framework integrations, and tools for building agent memory with Zep Cloud, the managed platform. The company has published a post announcing a new direction for its open-source strategy. Check their current documentation before assuming a specific self-hosting path.
Is Bilinc open source?
No. Bilinc is BUSL-1.1 and the public package is a cloud-only SDK, CLI, and MCP adapter. Local storage and private runtime internals are not shipped in the public package.
What does Bilinc do that a memory platform typically leaves to the application?
Bilinc treats the record around a write as part of the product: whether the write was verified, which run or operator produced it, what prior state it contradicted, and how to roll it back. Teams often build that layer themselves on top of a memory store.
Do both support MCP?
Bilinc ships a Cloud MCP adapter over stdio exposing commit_mem, recall, and status, with documented configurations for Claude Code, Codex, and Cursor. For Zep, check their current documentation for the supported integration surface.
How do I evaluate the two fairly?
Run the same agent workflow through both and check three things: can you recall the decision you need before a risky edit, can you tell who wrote a given piece of state, and can you undo a bad run without manual cleanup.
Place Bilinc against the rest of the stack
Still working out what you need?
The agent memory guides answer the underlying questions first — what agent memory is, where vector search stops, and how rollback and audit actually work.