I
Model generation
The raw material. Local by default through Ollama; any OpenAI-compatible endpoint by key. The model proposes — nothing more.
SilverSerf is the assay office for machine reasoning: a native Windows platform for Owen, a governed local analytical agent. Every answer arrives with its warrant — the exhibits it read, the boundaries it kept, and the trace of how it decided.
v1.0.0 · local-first · Ollama by default, any OpenAI-compatible API by key
Ordinary assistants collapse reasoning into one opaque act. SilverSerf splits it apart and keeps each stage on the record.
The raw material. Local by default through Ollama; any OpenAI-compatible endpoint by key. The model proposes — nothing more.
Every response is assayed: exact task, world finding, position finding, decisive evidence, authority scope, and the trigger that would revise it.
Durable character change passes through a commit gate. Corrections have standing, not sovereignty. Denied proposals stay visible in the ledger.
The operator sees what was decided, on what warrant, and can reopen the account. Repairs retract only what changed and preserve what survives.
Disclosure is not left to model compliance. After every answer, the runtime constructs the exhibit manifest itself — the message, every retrieved excerpt, every governed rule supplied — and displays it beneath the response. Owen may say which items were decisive. Owen cannot suppress the manifest.
Ask Owen to interpret a named document and retrieval binds to that document — or refuses. No exhibit, no interpretation. Fluent uncited book reports are withheld, not delivered.
Every response carries a structured audit: evidence, inference, uncertainty and rivals, revision triggers. Open the trace panel and read exactly how the answer was reached.
When an answer drops a distinction, name it. Owen retracts locally, preserves what survives, and the correction is staged for governance — never silently absorbed.
The hosted demo runs the same SilverSerf runtime in your browser — and Owen answers right away, no key or setup needed, with a small daily free allowance on a hosted model. For unlimited use, connect your own Ollama or any OpenAI-compatible API key in Settings, or upgrade to Pro. Every answer carries its exhibit manifest either way.
The full reasoning, audit, repair, and governance system is free when you bring your own compute. Pro adds managed convenience — never extra rigor withheld.
The full native experience: WPF desktop shell, managed Python and Node runtimes, automatic Ollama and gemma3:12b setup, SEED Bedrock substrate, and local-only loopback services.
SHA-256 34cef9d62455a8c3b10c68fdd8121e536baa178d6e8afd32a1a0786f388c9cc7
This build is Authenticode code-signed (publisher: David Carr); Windows may still show a SmartScreen reputation prompt while the signature builds reputation. Verify the checksum before running. The installer requires network access to fetch managed runtimes, Ollama, and the default model unless already cached.