ARKTOR Vault
Change the model without throwing away the knowledge your ARKTOR environment has built. Vault keeps approved project context local and reusable across chats, model changes and restarts, with source context and user control attached.
Who this product is for
Long-running AI workflows that need useful local context to survive chats, model changes and restarts without becoming opaque automatic memory.
The decision moment
Project facts, procedures or approved knowledge should persist across sessions but still remain reviewable, traceable and under user control.
Change the intelligence. Keep the working environment.
A team may use a frontier model today and move to a different provider or a local model later. Vault keeps approved project knowledge available to the next model instead of forcing the organisation to rebuild its context from scratch. The new model may be weaker or stronger, but the surrounding ARKTOR knowledge does not have to disappear with the switch.
Durable context with a clear boundary
- Retains approved knowledge across sessions and restarts
- Reuses memory across different supported models
- Keeps source and provenance context attached to durable knowledge
- Separates ordinary search evidence from promoted long-term memory
- Records outcome feedback without silently rewriting confidence
0.3.0-alpha.1 is more than a storage demo.
The current local alpha passed its retained 21/21 end-to-end memory lifecycle suite plus fresh build-quality gates. Cross-model proof also showed a stored procedure being reused after a fresh process by a different model.
Boundary: this proves the defined local memory workflows. It does not make every future extraction, ranking or multi-user scenario production-ready.
Remembering more is not automatically better.
Durable AI memory can amplify old mistakes if every observation is promoted without review. ARKTOR Vault keeps useful context reviewable and traceable instead of silently turning every observation into unquestioned truth.
Keep the project context when the session or model changes.
See the evidence-backed scenario: Keep AI project memory across sessions and models →
Verified local alpha, not a finished memory service.
- No general public installer or signed release yet
- Encryption-at-rest and multi-user authorization remain future release work
- Automatic extraction may be explored later, but untrusted memory is not silently promoted today
- Large-history retrieval should be measured before adding unnecessary indexing complexity
ARKTOR Vault 0.3.0-alpha.1 is a verified local ARKTOR family module. Public release readiness remains separate from engineering proof.