Kitaru vs Memharness

Kitaru and Memharness 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.

Kitaru

Kitaru wraps your existing agent SDK — PydanticAI, OpenAI Agents, Claude Agent SDK, or raw Python — and turns every model call, tool call, and intermediate step into a durable checkpoint. When you want to ask what would have happened with a cheaper model or a failed retriever, you replay from a specific checkpoint with one override. Nothing re-executes in production. The vendor's own benchmark shows 200 replayed executions on a cheaper model matching outputs in 192 of 200 cases at 84% lower cost. The ceiling appears when your agent's behavior depends on state that Kitaru's adapter doesn't intercept — external side effects or SDK internals the wrapper never sees won't be faithfully replayed.

Memharness

The core premise is storing facts, not strings, with two independent time axes: when something became true in the world and when the agent learned it — so querying past agent states is a real query, not archaeology through logs. Everything lives in a single SQLite file, which means the storage layer makes zero LLM or network calls and stays auditable. Recall combines hybrid vector search and full-text search with a source-staleness signal, so older or superseded sources rank down automatically. Where it breaks: the SQLite backend is a hard ceiling for teams expecting distributed writes or high-concurrency production deployments. Teams hitting that ceiling will need to treat memharness as a pattern to port, not a service to scale horizontally.

Attribute	Kitaru	Memharness
Pricing	Free	Free
Free trial	No	No
Open source	Yes	Yes
Has API	Yes	Yes
Self-hosted option	Yes	Yes
Platforms	Python, self-hosted cloud	SQLite, MCP
Pros	Replay from any named checkpoint with a single model or override argument, so testing a cheaper model against 200 real prior executions costs a fraction of re-running them live — the vendor reports 84% cost reduction in their own benchmark. Adapter-based instrumentation wraps the runner objects you already call (OpenAI Agents, Claude Agent SDK, PydanticAI, raw Python), so you are not rewriting agent logic to get checkpointing — the diff between instrumented and uninstrumented code is a class swap. Tool failure simulation via override parameters (`lookup_mode=timeout`, `overrides={checkpoint.lookup_order: stale_order}`) means you can reproduce edge cases that only appeared in one production run without manufacturing a synthetic fixture that may not match real call structure. Durable checkpoints support human approval gates mid-execution, so a compliance or review step can pause the agent at a named checkpoint and wait for a sign-off before continuing — without rebuilding the agent's control flow. Apache 2.0 open source with a self-hosted path, so the checkpoint store and replay engine stay inside your infrastructure boundary — no production trace data leaves your environment unless you opt into ZenML Pro.	Bi-temporal storage tracks both world-time and agent-learn-time independently, so you can reconstruct exactly what the agent believed at any past moment — which means post-incident reviews and compliance audits have an actual record to query instead of inferring from logs. Provenance-scoped deletion lets you remove all facts derived from a specific source in one operation, so GDPR takedown requests or source revocations do not require a full memory wipe that destroys unrelated facts. The storage layer makes zero LLM or network calls, so memory reads and writes have no latency dependency on external APIs and no token cost — which means memory operations do not blow your inference budget. Hybrid vector-plus-full-text recall with a built-in staleness signal means older or superseded sources rank lower automatically, so the agent surfaces the most current relevant facts without you writing custom re-ranking logic. MCP exposure and a self-hosted SQLite backend mean the tool drops into any agent stack that speaks MCP without requiring a separate managed service, so you retain full data ownership and avoid a vendor dependency in the memory layer.
Cons	Replay fidelity breaks when agent behavior depends on state the adapter never intercepted — external database reads that changed between the original run and the replay, webhook side effects, or SDK internals the wrapper sits outside of will produce divergent results, and the what-if conclusion becomes unreliable. Teams working around this add manual checkpointing at additional call sites, which means maintaining instrumentation that grows with the agent's surface area. Adapter coverage is scoped to PydanticAI, OpenAI Agents SDK, Claude Agent SDK, and raw Python — teams running other frameworks or heavily customized SDK forks have no adapter and must write their own wrapper against Kitaru's primitives, which the docs describe as 'agent runtime primitives and APIs' without a pre-built path. Teams on LangChain or LlamaIndex agent stacks have no drop-in adapter and will evaluate alternatives that instrument at the framework level. The replay model assumes the original run was fully checkpointed — if an agent crashed before a checkpoint was written, there is nothing to replay from. Recovery from mid-run crashes depends on checkpoint granularity configured at instrumentation time, not retroactively. Teams that instrument coarsely discover this the first time they need to replay a run that ended partway through a tool chain.	SQLite is a single-writer database: the moment two agent processes attempt concurrent writes — a parallelized pipeline, a multi-worker deployment, any architecture where more than one process holds the file — writes will collide or block. Teams with concurrent-write requirements either serialize all memory operations through a single process (adding a bottleneck) or abandon memharness for a Postgres- or Redis-backed alternative. The project has 2 stars and 1 fork on GitHub at time of curation, with 19 commits and no open issues, which means community-sourced debugging, third-party integrations, and production war stories are essentially nonexistent. Teams that hit an edge case are reading the source, not a Stack Overflow thread. There is no built-in access control or multi-tenant isolation: if multiple agents or users share the same SQLite file, provenance-scoped deletion could become a liability rather than a feature — one delete call wipes facts for every tenant who learned from that source. Teams building multi-user applications will need to implement per-user database files or a sharding layer before going to production.

Bottom line

Kitaru and Memharness 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 Kitaru and Memharness?

Kitaru is Free and open source, while Memharness is Free and open source. Compare pricing, free trial, API, platforms, and pros/cons in the table above on AIDiveForge.

Is Kitaru better than Memharness?

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.