Provenance and knowledge records
How to record the origins and development of knowledge, including human and AI contributions. Earlier bitknowledge ID and durable-record experiments provide background for this research.
Engineering at bitknowledge
Getting a model to respond is only part of the job. Bitknowledge works on the infrastructure and integrations that make AI usable within an organisation, whether the model runs locally or through a commercial service.
Testing before committing
A small test can answer questions that a specification alone cannot: whether a model handles the intended task, whether response times are acceptable, or whether the available hardware is sufficient.
Bitknowledge uses prototypes and practical evaluation where they help resolve uncertainty. Results, limitations and trade-offs should be recorded so there is a clear basis for deciding what to build next.
Looking after the whole system
On local infrastructure, the details include Linux services, GPU drivers and CUDA compatibility, model runtimes, storage capacity and inference performance. With hosted model APIs, the work includes authentication, integration, usage monitoring and handling provider limits or outages.
In either case, data flows and access controls need to be understood. Logs should help diagnose problems without unnecessarily retaining sensitive information. Costs, resource use and service health need to be visible to the people responsible.
Configuration, dependencies, upgrade procedures and recovery arrangements should be documented. Where practical, components should be replaceable without rebuilding the whole system.
Open-source software and documentation
Open-source tools can make it easier to inspect behaviour, diagnose faults and maintain a system over time. They also give people something concrete to learn from and adapt. That is useful only when the tools fit the task and can be supported; commercial software and hosted services can be the right choice too.
Good documentation connects the software to the actual installation: how it is configured, why decisions were made, what depends on it and how to change it safely. Practical walkthroughs help transfer that knowledge.
Bitknowledge intends to publish useful open-source tools where project work reveals recurring problems worth solving more generally.
Exploratory work
Alongside the infrastructure work, bitknowledge has longer-term technical and research interests in how AI systems handle knowledge and how people retain control over them.
User-owned AI systems
Infrastructure, memory and security arrangements that give people control over personal AI services and the context those services accumulate.
Inscriptions and provenance records
The inscriptions experiment uses structured metadata and content-addressed records to preserve published versions. The public index currently contains the current and archived founder messages; historical source material remains preserved in the repository.
View the provenance records