Enju vs Two-tier-memory

Enju and Two-tier-memory are both agent frameworks tracked by AIDiveForge. Below is a side-by-side comparison of pricing, capabilities, platforms, and ownership — sourced from each tool's live website and verified before publishing.

Enju

Orbit structures agent work into discrete, dependency-ordered loops: one task per run, deterministic validation gates, and four output artifacts that record exactly what the agent returned, how the run scored against a rubric, and what should happen next. The demo runs without an API key, which means you can evaluate the harness itself before spending a single token. Where it gets constrained: Orbit is a harness, not a scheduler — it does not autonomously drive through a backlog or retry failed orbits on its own. Teams wiring it into CI pipelines write the outer loop themselves.

Two-tier-memory

The library implements what the repo calls the 'two-tier fix': structured storage in a local SQLite database, with semantic or keyword queries pulling back only the relevant rows instead of the entire memory corpus. The core workflow is a single Python file and a SQL schema — add a memory, query a memory, done. It runs entirely on-device with no external API calls. The wall you hit is expressiveness: the schema is fixed, so teams with complex memory taxonomies end up forking the schema or layering their own abstraction on top. At that point you are maintaining a fork.

Attribute	Enju	Two-tier-memory
Pricing	Free	Free
Free trial	No	No
Open source	Yes	Yes
Has API	No	No
Self-hosted option	Yes	Yes
Platforms	Platform-agnostic (Python); local or remote execution	Python, SQLite
Pros	Agent-neutral adapter contract, so you can run Claude and Codex against the same task definition and compare structured evaluation artifacts instead of arguing over impressions. Validation gates (tests, lint, type checks) block task completion until checks pass, which means agent output that merely looks correct cannot close an orbit and cannot reach your branch. Dependency-aware backlog selection keeps each run scoped to a single, well-bounded task, so you avoid the compounding failures that come from an agent chaining through multiple ambiguous steps at once. Mock-mode replay demo requires no API key, so you can evaluate Orbit's harness behavior and artifact output without spending tokens or standing up external credentials. MIT licensed and self-hostable, which means no vendor dependency on the validation layer for a security-sensitive or air-gapped environment.	Queries the SQLite store for only the relevant memory entries rather than loading the full history into context, so the agent's effective memory scales with the size of the database rather than the size of the context window. Entirely local and dependency-light — no API keys, no network calls, no managed service — which means the memory layer cannot go down because a third-party endpoint is unavailable. MIT license with full source in a single Python file, so you can read exactly what happens to your stored data and modify the retrieval logic without waiting on a vendor. CLI-driven interface means you can add or query memories from shell scripts, editor plugins, or agent tool calls without importing a framework. Persistent across sessions by default via SQLite, so a solved problem recorded in one session is available in every subsequent session without any additional configuration.
Cons	Orbit does not drive its own retry or backlog progression loop — when an orbit fails validation, a human or an external script decides what runs next. Teams expecting autonomous multi-task execution will write a significant orchestration layer on top of the harness before it matches that expectation. There is no API surface and no native CI integration out of the box. Connecting Orbit to a GitHub Actions pipeline or a merge queue requires an adapter the team authors; the docs describe this as a contribution pattern, not a built-in feature. The harness is scoped to coding agents that speak a JSON CLI contract. Teams already invested in a coding agent that does not expose a structured CLI output format will hit an integration wall immediately and either write a translation shim or move to a validation approach their agent already supports natively.	The schema ships with a fixed structure targeting solved coding problems and project decisions. Teams whose memory needs include different record types — hierarchical documentation, multi-entity relationships, or domain-specific metadata — hit the schema ceiling immediately and must fork and migrate, at which point they own all future schema evolution. There is no server, no sync layer, and no multi-agent access model. A team with more than one agent process, or a developer working across multiple machines, gets no shared state — each environment has its own isolated database, and keeping them consistent is a manual problem. At the point where a team needs semantic vector search rather than keyword or structured queries — typical once the memory corpus grows large and queries become fuzzy — this library provides no embedding or vector retrieval path. That is the condition under which teams move to a dedicated vector database or a memory framework like Mem0 instead.

Bottom line

Enju and Two-tier-memory are closely matched on pricing model, openness, and API availability — pick by feature set and platform support in the table above.

Frequently asked questions

What is the difference between Enju and Two-tier-memory?

Enju is Free and open source, while Two-tier-memory is Free and open source. Compare pricing, free trial, API, platforms, and pros/cons in the table above on AIDiveForge.

Is Enju better than Two-tier-memory?

It depends on your workflow. Use the side-by-side attributes (pricing, open source, API, self-hosted, platforms) to decide. AIDiveForge does not rank a universal winner — we publish verified facts so you can choose.