SERVE

Top-K always returns K.
Memoth can return nothing.

That is the failure mode: an agent asks a question your memory has no answer to, and a similarity search hands it the ten nearest paragraphs anyway. Memoth serves through persona lanes, promotes with IDF weighting, and holds an abstention floor. Below the floor the agent is served nothing — and told that is what happened.

EXTRACT · GOVERN · SERVE

What Memoth is

A store, a ledger, and a write-back loop. One layer under every agent.

MEMORY STORE
EPISODICpayments failover — what broke
SEMANTICtopology: active-active
PROCEDURALrollback runbook · 6 steps

Typed fragments, not chunks — what a mind would keep.

GOVERNANCE
source: Arch p.12 · verified
merge recorded · ledger #4821
ACL: devops → read

Every memory carries its source; every change hits a ledger.

PERSISTENT LAYER
learned: retry window 30s
written back · 2m ago
week-2 recall +18%

What agents learn writes back. Week two beats week one.

Memoth.aiThe memory layer

How it actually works

Three mechanisms between a question and a context window.

PERSONA LANES

Each agent is bound to a role card — persona, domain, scope tags — and that card is the lane it can be served from. An agent bound to no card is served everything, which is literally the promotion semantics, so Memoth flags it rather than letting it pass as configured.

Only what fits its persona, domain and task.

IDF PROMOTION

Candidates inside the lane are promoted with inverse-document-frequency weighting, so a term that appears in every document weighs less than one that appears in a few. Each served fragment comes back with the score and the lane that promoted it, not just the text.

Promoted at the moment it helps, rather than stuffed into the front of every prompt.

ABSTENTION FLOOR

Below the floor, nothing is promoted. The agent receives an abstention with its reason instead of the nearest paragraph, and the reason names the source that would have had to exist for an answer to be possible.

What agents learn while working is written back, so week two is smarter than week one.

Measured, not promised

We publish the numbers behind every claim.

37/40
Answers correct on our document-QA suite, held to a zero-hallucination CI bar: one made-up answer fails the build.
40% → 80%
Task success with conditioned memory on CAMP-Bench, with hallucination cut to a third.
CAMP-Bench
Our benchmark for promotion: the right memory, to the right agent, at the right time, per domain. Every release is judged on it.

Both figures are reproducible on a frozen fixture, not third-party-validated. We run the same suites on your corpus during a pilot and hand you the table.

WHAT MEMOTH SERVED

Every answer shows its working set.

Both panels below are the product's serving surface — the fragments in them are examples, not a live tenant. The second one is not an error state. It is the same surface reporting that nothing cleared the floor.

WHAT MEMOTH SERVED

FROM YOUR DOCUMENTS (3 fragments)product UI · illustrative

devops-agent asked: “what is the cluster topology for payments?”

  • payments runs single-region in ap-south-1

    DevOps / Architecturesemanticconf 0.94verbatim
  • failover rehearsal, step 3: drain the read replica first

    DevOps / Runbooksproceduralconf 0.88grounded paraphrase
  • the 2026-04 replica rollout was reverted the same week

    DevOps / Incidentsepisodicconf 0.71unverified

Abstained

NOT COVEREDproduct UI · illustrative

devops-agent asked: “what is the on-call rota for the billing team?”

When nothing fits, the agent gets nothing.

No fragment in this agent's lane cleared the abstention floor, so none was promoted. The answer names the source that would have had to exist and generates nothing to cover the gap.

lane: devops0 fragments promoted

Abstention rate is a per-agent number on the dashboard, reported as governance working rather than as a failure. An agent that never abstains is not a better agent — it is an agent whose floor is too low.

Design partners

We're choosing a handful of teams to build with.

What you get

The memory layer under your agents early, tuned to your domain first. We measure token cost, task success and served-stale rate, and hand you the numbers. If they don't show up, you walk with a free audit.

Founder-direct iteration. Your corpus never leaves your environment.