Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
Provider inventory and reasoning intent
Owner decision: 2026-09-08 12:33 +08. This supersedes the Source tier editor and the resolver's exact-declaration forwarding rule, not model-menu selection, subscription OAuth, credential replacement, or routing policy.
Contract
- Provider inventory exposes model identity and inventory actions, not a second reasoning configuration surface. This holds for every metadata provenance and both discovered and manual models. Remove the empty advanced toggle as well.
- Manual creation needs only the model ID. It sends an empty capability list through the existing API and does not fetch upstream inventory.
- Removing controls does not rewrite stored capability metadata, remove models, alter existing routes or change another consumer's compatibility API.
- The shared resolver forwards the caller's reasoning request unchanged to the managed adapter for each exact Source/model attempt. Capability declarations are not an allowlist. Fallback and credential retry start from the same caller intent, preserving protocol, headers and non-effort fields.
- Protocol errors keep their existing terminal behavior. No silent fallback or synthesized effort is added. Historical stripping telemetry remains readable; new resolver attempts do not claim stripping that no longer happens.
- Amendment (planned, capability model §3.4):
the effort is still forwarded unchanged, but a hop with explicit negative
evidence for that effort is skipped for the request, and a request that no
hop can serve fails before invocation with
effort_unsupported.
Proxy engine boundary
This change proves forwarding through Avibe's gateway and resolver to its
adapter. It does not claim end-to-end preservation through the pinned proxy
engine's translators. CLIProxyAPI v7.2.149 (2a6b87aca083a5bf498ac1f68a1b636c500d7aaa)
has a second capability policy: internal/modelconfig/model_info.go marks
configured API-key models non-user-defined, and internal/thinking/apply.go
can strip thinking for models with no registered capability or validate a
configured level list. Removing the Avibe filter alone cannot close that gap.
Source model registration is intentionally unchanged here: simply omitting its thinking metadata can make the engine strip more parameters, not fewer. A separate engine policy/release change needs verification with a local mock upstream across protocol families before an end-to-end guarantee is possible. No live credentials or operational Avibe instance are required or used.
Evidence
- Unit: every valid provenance shape, empty/mismatched capability metadata, direct and nested effort, Anthropic thinking, and absence of reasoning intent.
- Scenario: MH-EFFORT-001 and D10 traverse the HTTP turn gateway and exact adapter attempts; existing API capability ownership remains covered by B10.
- UI: inventory and manual-add controls independent of capability metadata; existing provider management, mutation settlement, OAuth and replacement.
- Browser: isolated English/Chinese desktop/mobile provider fixture, with no live API access. Full local Incus and proxy-engine wire verification remain separate evidence layers, not inferred from a fake adapter.