Observant, not autonomous
Models are expected to inspect context, use tools, and surface their reasoning. Jaguar treats good tool use and memory activation as observable behavior, not magic.
Jaguar is the human-in-the-loop layer that gives Wanatux OS its shape: persistent memory, approval-gated actions, and audit trails that let different models work inside the same environment without pretending autonomy they do not have.
Jaguar is the operating stance that keeps the system useful without turning it into a black box. It narrows the contract between human, model, memory, and tools.
Models are expected to inspect context, use tools, and surface their reasoning. Jaguar treats good tool use and memory activation as observable behavior, not magic.
Mutations are guarded behind confirm and dry-run paths. The system helps propose, stage, and explain work, but it does not erase operator judgment.
Jaguar is the public-facing description of a personal operating environment. It exists to explain the system honestly, not to market a fictional platform.
The loop is simple: retrieve context, propose a move, approve explicitly, then record what happened so the next session starts from a stronger place.
Models query the environment first: memory, tools, data stores, and prior runs.
They stage a next action instead of improvising from memory alone.
Writes and destructive work stay behind explicit operator approval gates.
Results, decisions, and corrections are written back for future recall and audit.
Jaguar treats context as infrastructure. Retrieval spans live session artifacts and long-term knowledge so models can look things up rather than reconstruct them unreliably.
The system can stage calendar actions, todo updates, file operations, and maintenance flows, but Jaguar keeps the decision boundary visible.
These are not roadmap claims. They are already running surfaces inside the lab and they are the reason Jaguar is worth naming at all.
Memory recall spans live devloop artifacts and indexed knowledge so models can recover prior decisions, documents, and state snapshots in one pass.
Due-dated tasks can be reviewed, staged, mirrored into Google Calendar, and audited without collapsing the approval boundary.
The environment already runs daily quick-wins summaries, Sigil health checks, and other operational observers that keep the system from going stale.
GPT, Claude, and local models can work against the same memory and tool layer, which makes behavior comparable instead of anecdotal.
Jaguar is the house. The Sigil page, project surface, and OSF research record are the proof that the house is being used for real work.
The Sigil page shows the measurable side of the epistemic layer: observed model behavior, logging adoption, and structured recall.
Open Sigil →The public paper and appendix components provide the formal research frame for the same environment described here.
Read the record →Selected tools and servers from the lab are published with sanitized configs and working documentation.
View projects →Quiet authority.
Jaguar marks the shift from scattered experimentation to deliberate authorship. The system does not need to look loud to hold its boundary. It just needs to remain coherent under pressure.