Is your feature request related to a problem?
AgentOS leans hard into cognitive memory, which is exactly why I'd love to make a strong long-term memory backend a first-class option. Today, plugging in an external memory system that organizes durable facts (rather than raw history) and does consolidation/recall server-side takes manual wiring.
Describe the solution you'd like
A Hindsight integration for AgentOS. Hindsight is an open-source long-term agent memory system: it organizes memory into world facts, experience facts, and consolidated mental models, with multi-strategy recall (semantic + BM25 + graph + temporal) and a reflect operation.
Shape would mirror how AgentOS already models memory — recall relevant memories into context before a model call, and retain new memories after a turn (entity extraction + consolidation handled by Hindsight). I'm the maintainer and happy to build and maintain it.
Describe alternatives you've considered
- Implementing against AgentOS's native memory interface directly (tighter, first-party).
- An external wrapper package that layers Hindsight on top of agent config (lighter touch).
Happy to go whichever way fits your architecture.
Additional context
Prior art: Hindsight already ships as a first-party memory provider in NousResearch's Hermes agent, alongside honcho, mem0, supermemory, and byterover. Mainly want to gauge interest and preferred location before opening a PR.
Is your feature request related to a problem?
AgentOS leans hard into cognitive memory, which is exactly why I'd love to make a strong long-term memory backend a first-class option. Today, plugging in an external memory system that organizes durable facts (rather than raw history) and does consolidation/recall server-side takes manual wiring.
Describe the solution you'd like
A Hindsight integration for AgentOS. Hindsight is an open-source long-term agent memory system: it organizes memory into world facts, experience facts, and consolidated mental models, with multi-strategy recall (semantic + BM25 + graph + temporal) and a
reflectoperation.Shape would mirror how AgentOS already models memory — recall relevant memories into context before a model call, and retain new memories after a turn (entity extraction + consolidation handled by Hindsight). I'm the maintainer and happy to build and maintain it.
Describe alternatives you've considered
Happy to go whichever way fits your architecture.
Additional context
Prior art: Hindsight already ships as a first-party memory provider in NousResearch's Hermes agent, alongside honcho, mem0, supermemory, and byterover. Mainly want to gauge interest and preferred location before opening a PR.