Registration of interest open
Sydney
- Date
- TBC
- Venue
- TBC
- Host
- TBC
Most agent memory today is just replaying the whole conversation history back into the model. That is not memory, it is a workaround.
For AI engineering, platform, data and technology leaders.
An agent that remembers everything remembers nothing useful. Replaying an entire history back into the context window is expensive, slow, and quietly wrong, because contradictions from six turns ago arrive with the same weight as the fact that replaced them.
That is the starting point for this series. Not a product session, and not a briefing. A table of senior people working through what it means to treat memory as infrastructure that is actively maintained, rather than a transcript that is endlessly re-read.
Six cities across APAC, between October and November 2026. Each table runs under the Chatham House Rule.
Weaviate press materials, February and June 2026. Figures are reproduced from the published releases and are pending client verification.
Each table works through the same three questions, shaped by what the room brings to them.
Every team hits the same wall from a different direction: context windows that cost more than the inference, latency that creeps past what a user will tolerate, or an agent that contradicts itself because two versions of the truth are both still in the transcript.
// tension: a longer context window is not a memory strategyIf facts are reconciled asynchronously rather than appended forever, something has to decide which version wins. This session looks at who owns that reconciliation logic, how it is tested, and what it costs to run alongside an existing retrieval stack.
// differentiator: memory reconciled asynchronously, not accumulatedDeveloper-led adoption and enterprise procurement pull in opposite directions. The table works through how teams have taken an open-source-first choice through security review without losing the flexibility that made it attractive.
// context: developer-led adoption meeting enterprise governanceEach city is its own table with its own room, confirmed independently. Registration of interest is open now, with dates and venues confirmed city by city.
Memory should be actively maintained, not left to accumulate.
Weaviate Engram general availability announcement, June 2026
Published Weaviate material behind the questions on the table. Share your details once to unlock all three.
Tell us which city suits you and we will come back to confirm your spot.
Weaviate began as SeMI Technologies and took the name of its own vector database in January 2023. The company is remote-first and open-source-first, and the database itself remains freely installable, which is how most of its adoption starts.
The platform now covers storage, retrieval and orchestration for generative AI, with Engram extending it into managed agent memory. Its identity was deliberately built against the isometric-grid sameness of the vector search category.
Innovatus Media is a Sydney based B2B events agency that runs invitation-only executive roundtable series across APAC, EMEA and North America.
We convene senior decision makers for closed door conversations on the problems they are actually working on, and produce every element of the series end to end, from the room and the table to the follow up.
Connect for more details on the series.