The landing page now follows the same public canon as the white paper, technical appendix, roadmap, and benchmark surface.
id-mem is a memory runtime for agents and teams: canonical ingest, shared world state, perspective-aware retrieval, authority handling, temporal state, and an explicit MCP control surface.
The landing page now points directly at the current canonical materials. Thesis, proof, benchmarks, and roadmap each have their own place.
The product thesis. What id-mem is, how the runtime is meant to be understood, and the current public scope.
The proof layer. Research grounding, mechanism, claim-by-claim status, and the benchmark logic behind the public story.
The stress surface. Adversarial runtime tests, divergence gates, determinism runs, and cross-provider comparisons.
What comes next. Future intent stays separate from current proof so the public story does not blur shipped versus planned.
The important shift is not storing different truths per agent. It is storing one world and ranking that world differently depending on who is asking.
“What should we prioritize now?”
Pipeline predictability, budget exposure, and owner clarity move to the top. The same world state is read through a commercial and decision-heavy lens.
“What should we prioritize now?”
Client campaigns, content cadence, and audience response rise first. The same world state is read through brand, narrative, and delivery resonance.
“What should we prioritize now?”
Revenue concentration, operating risk, and near-term confidence become dominant. The same memories are ranked through return, risk, and financial clarity.
This is the live idea behind the appendix and benchmark story: perspective is not random drift. It is repeatable ranking divergence over the same shared memory.
The public wording is now aligned with the canonical documents. The landing page should describe the runtime that exists, not a broader marketing abstraction.
Slack, API events, MCP flows, file imports, and plugin hooks enter through one canonical path with provenance and idempotent handling.
Write time builds one workspace memory layer. The stored world is not duplicated per agent and does not depend on who asks later.
Retrieval applies agent-specific priorities, worldview, and talent signals when ranking relevance. Perspective changes order, not the underlying truth.
The runtime treats supersession, current versus historical state, and source authority as first-class semantics instead of flat text similarity.
These are the suites worth reading if you want to know whether id-mem is doing more than generic retrieval. They are now aligned with the canonical appendix instead of being hidden behind internal docs.
Same workspace, same corpus, only the agent lens changes. This is the cleanest proof that perspective changes retrieval instead of prompt styling.
The same agent sees the same query five times. The current stored run shows stable repeatability while different agents still separate sharply.
Retrieval quality and agent differentiation are measured together on one harness, because relevance without perspective is not enough.
The same corpus is queried across six providers and four perspectives. id-mem diverges in a controlled way where generic memory systems stay nearly flat.
Authority, temporal change, perspective divergence, and determinism are rerun on the VPS, not only in offline benchmark JSON.
API and MCP are checked against the same memory actions so the control surface stays explicit across product entry points.
Public benchmark language is deliberate: live runtime proof is stronger than repo-only implementation, and strict gates are kept visible even when they reveal remaining work.
The canonical public story is now split cleanly between shipped behavior, proof surfaces, and roadmap intent.
The landing page is now a gateway into the public canon, not a parallel narrative. White paper explains the thesis. Appendix explains the mechanism and proof. Benchmarks show stress behavior. Roadmap explains what comes next.
If you want the thesis, start with the white paper. If you want the proof layer, go straight to the appendix and benchmarks.