@openllmsh/sdk 0.0.0-stage → 2.10.12-alpha.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/ARCHITECTURE.md +43 -0
- package/LICENSE +202 -0
- package/README.md +142 -2
- package/assets/openllm-light.svg +18 -0
- package/assets/openllm.svg +18 -0
- package/client/chat.ts +140 -0
- package/client/devices.ts +269 -0
- package/client/http.ts +72 -0
- package/client/index.ts +129 -0
- package/client/relay-mux.ts +547 -0
- package/client/types.ts +51 -0
- package/client/vault.ts +239 -0
- package/connect/chat.ts +58 -0
- package/connect/client.ts +240 -0
- package/connect/config.ts +47 -0
- package/connect/errors.ts +121 -0
- package/connect/index.ts +23 -0
- package/connect/oauth.ts +265 -0
- package/connect/routes.ts +234 -0
- package/connect/store.ts +107 -0
- package/mcp/handle.ts +24 -0
- package/mcp/index.ts +76 -0
- package/mcp/server.ts +319 -0
- package/mcp/tools/marketplace.ts +439 -0
- package/mcp/tools/memory-client.ts +297 -0
- package/mcp/tools/memory-config.ts +32 -0
- package/mcp/tools/memory.ts +369 -0
- package/mcp/tools/model-priority.ts +88 -0
- package/mcp/tools/openllm.ts +883 -0
- package/mcp/types.ts +28 -0
- package/openapi/call.ts +90 -0
- package/openapi/index.ts +8 -0
- package/openapi/openapi.json +80213 -0
- package/openapi/operations.ts +2075 -0
- package/package.json +39 -4
- package/protocol/CONTRIBUTING.md +21 -0
- package/protocol/README.md +134 -0
- package/protocol/access.ts +186 -0
- package/protocol/agent-prompt-text.ts +32 -0
- package/protocol/agent.ts +991 -0
- package/protocol/api-key-syntax.ts +105 -0
- package/protocol/assets/openllm-light.svg +18 -0
- package/protocol/assets/openllm.svg +18 -0
- package/protocol/audio.ts +29 -0
- package/protocol/auth.ts +142 -0
- package/protocol/changelog.ts +155 -0
- package/protocol/chat.ts +458 -0
- package/protocol/client-registry.ts +356 -0
- package/protocol/config.ts +345 -0
- package/protocol/control-channel.ts +115 -0
- package/protocol/cooldown-reason.ts +78 -0
- package/protocol/cost.ts +10 -0
- package/protocol/daemon-port.ts +31 -0
- package/protocol/daemon.ts +1524 -0
- package/protocol/default-chains.ts +66 -0
- package/protocol/device-state.ts +85 -0
- package/protocol/doctor-report-local.ts +443 -0
- package/protocol/doctor-report-policy.ts +119 -0
- package/protocol/doctor-report.ts +947 -0
- package/protocol/errors.ts +14 -0
- package/protocol/execution-selection.ts +141 -0
- package/protocol/hop-grant.ts +33 -0
- package/protocol/image-edit-parse.ts +263 -0
- package/protocol/images.ts +298 -0
- package/protocol/index.ts +74 -0
- package/protocol/inference-headers.ts +3 -0
- package/protocol/keepalive.ts +69 -0
- package/protocol/local-auth.ts +300 -0
- package/protocol/local-device-control.ts +133 -0
- package/protocol/local-provider.ts +288 -0
- package/protocol/local-session.ts +118 -0
- package/protocol/marketplace.ts +187 -0
- package/protocol/media-attempt.ts +81 -0
- package/protocol/media-bytes.ts +28 -0
- package/protocol/media-default.ts +357 -0
- package/protocol/media-dimensions.ts +132 -0
- package/protocol/media-input.ts +61 -0
- package/protocol/media-model.ts +142 -0
- package/protocol/media-option-issue.ts +90 -0
- package/protocol/media-types.ts +70 -0
- package/protocol/model-match.ts +1051 -0
- package/protocol/model-settings.ts +327 -0
- package/protocol/models.ts +767 -0
- package/protocol/mux.ts +722 -0
- package/protocol/plan-surface.ts +59 -0
- package/protocol/provider-models.ts +16 -0
- package/protocol/provider-status.ts +206 -0
- package/protocol/provider-tier.ts +67 -0
- package/protocol/provider-usage.ts +263 -0
- package/protocol/providers.ts +901 -0
- package/protocol/public-usage-index.ts +354 -0
- package/protocol/realtime-generated.ts +31 -0
- package/protocol/realtime.ts +324 -0
- package/protocol/registry-pointer.ts +118 -0
- package/protocol/relay.ts +702 -0
- package/protocol/replay-session.ts +11 -0
- package/protocol/responses.ts +280 -0
- package/protocol/semver.ts +182 -0
- package/protocol/setup-mode.ts +22 -0
- package/protocol/sha256-hex.ts +98 -0
- package/protocol/stats.ts +406 -0
- package/protocol/streaming.ts +19 -0
- package/protocol/subscription-daemon-floors.ts +33 -0
- package/protocol/subscription-provider-compat.ts +104 -0
- package/protocol/subscription-provider.ts +32 -0
- package/protocol/team-grant.ts +17 -0
- package/protocol/team-shareable.ts +61 -0
- package/protocol/transcription-encoding.ts +76 -0
- package/protocol/url-locality-dns.ts +37 -0
- package/protocol/url-locality.ts +288 -0
- package/protocol/usage-snapshots.ts +9 -0
- package/protocol/vendor-cli-install.ts +29 -0
- package/protocol/video-job-plan.ts +50 -0
- package/protocol/videos.ts +227 -0
- package/tsconfig.json +14 -0
- package/tunnel/ARCHITECTURE.md +111 -0
- package/tunnel/LICENSE +33 -0
- package/tunnel/README.md +45 -0
- package/tunnel/assets/openllm-light.svg +18 -0
- package/tunnel/assets/openllm.svg +18 -0
- package/tunnel/channel-envelope.ts +139 -0
- package/tunnel/codec.ts +191 -0
- package/tunnel/device-anchor.ts +229 -0
- package/tunnel/device-control.ts +118 -0
- package/tunnel/device-grant.ts +173 -0
- package/tunnel/fleet.ts +77 -0
- package/tunnel/index.ts +22 -0
- package/tunnel/mux.ts +580 -0
- package/tunnel/peer-link.ts +390 -0
- package/tunnel/rtc-auth.ts +415 -0
- package/tunnel/rtc-duplex.ts +187 -0
- package/tunnel/rtc-ice.ts +136 -0
- package/tunnel/secure.ts +1298 -0
- package/tunnel/stream-reset-error.ts +14 -0
- package/tunnel/streams.ts +1430 -0
- package/tunnel/team-grant.ts +302 -0
- package/tunnel/wire-bytes.ts +102 -0
- package/vault/ARCHITECTURE.md +470 -0
- package/vault/LICENSE +33 -0
- package/vault/api-key.ts +190 -0
- package/vault/argon2.ts +106 -0
- package/vault/cli-auth-seal.ts +157 -0
- package/vault/cred-seal.ts +146 -0
- package/vault/credential.ts +55 -0
- package/vault/crypto.ts +210 -0
- package/vault/daemon-attest.ts +229 -0
- package/vault/derive.ts +113 -0
- package/vault/device-access.ts +87 -0
- package/vault/device-anchor.ts +185 -0
- package/vault/device-control.ts +174 -0
- package/vault/handoff.ts +396 -0
- package/vault/index.ts +21 -0
- package/vault/passkey.ts +243 -0
- package/vault/phrase.ts +65 -0
- package/vault/pin.ts +153 -0
- package/vault/public-key-seal.ts +131 -0
- package/vault/seed.ts +272 -0
- package/vault/team.ts +355 -0
- package/vault/vault-item.ts +131 -0
- package/vault/vault-migrate-digest.ts +29 -0
- package/vault/wrapped.ts +77 -0
- package/wire/ARCHITECTURE.md +90 -0
- package/wire/CONTRIBUTING.md +19 -0
- package/wire/LICENSE +33 -0
- package/wire/README.md +51 -0
- package/wire/adapters/messages/anthropic-map.ts +37 -0
- package/wire/adapters/messages/index.ts +7 -0
- package/wire/adapters/messages/reasoning-from-items.ts +38 -0
- package/wire/adapters/messages/reasoning-signature.ts +243 -0
- package/wire/adapters/messages/request.ts +574 -0
- package/wire/adapters/messages/response.ts +215 -0
- package/wire/adapters/messages/streaming.ts +1027 -0
- package/wire/adapters/messages/usage.ts +43 -0
- package/wire/adapters/responses/index.ts +3 -0
- package/wire/adapters/responses/request.ts +395 -0
- package/wire/adapters/responses/response.ts +88 -0
- package/wire/adapters/responses/streaming.ts +344 -0
- package/wire/assets/openllm-light.svg +18 -0
- package/wire/assets/openllm.svg +18 -0
- package/wire/features/account-rank.ts +166 -0
- package/wire/features/compaction/compact-request.ts +862 -0
- package/wire/features/compaction/compaction-text.ts +43 -0
- package/wire/features/context-demote.ts +47 -0
- package/wire/features/context-overflow-strategy.ts +26 -0
- package/wire/features/context-skip.ts +170 -0
- package/wire/features/conversation-affinity.ts +144 -0
- package/wire/features/max-tokens-backfill.ts +141 -0
- package/wire/features/model-caps.ts +53 -0
- package/wire/features/model-settings.ts +24 -0
- package/wire/features/output-limit-clamp.ts +64 -0
- package/wire/features/quota-gate.ts +85 -0
- package/wire/index.ts +44 -0
- package/wire/lib/attempt-outcome.ts +43 -0
- package/wire/lib/canonical/content-part.ts +38 -0
- package/wire/lib/canonical/count-tokens-capability.ts +138 -0
- package/wire/lib/canonical/encoding-select.ts +82 -0
- package/wire/lib/canonical/image-signature.ts +202 -0
- package/wire/lib/canonical/image-sse.ts +216 -0
- package/wire/lib/canonical/message.ts +43 -0
- package/wire/lib/canonical/token-estimate.ts +266 -0
- package/wire/lib/encrypted-content.ts +66 -0
- package/wire/lib/error-class.ts +154 -0
- package/wire/lib/error-envelope.ts +56 -0
- package/wire/lib/forwarded-headers.ts +180 -0
- package/wire/lib/provider-error-rules.ts +48 -0
- package/wire/lib/refusal.ts +38 -0
- package/wire/lib/responses-apply-patch.ts +104 -0
- package/wire/lib/responses-namespace.ts +53 -0
- package/wire/lib/secret-scrub.ts +101 -0
- package/wire/lib/streaming/accumulate.ts +221 -0
- package/wire/lib/streaming/finish-reason.ts +33 -0
- package/wire/lib/streaming/peek.ts +154 -0
- package/wire/lib/streaming/provider-decode.ts +259 -0
- package/wire/lib/streaming/response-stream.ts +71 -0
- package/wire/lib/streaming/sse.ts +699 -0
- package/wire/lib/streaming/strip-tool-calls.ts +95 -0
- package/wire/lib/streaming/upstream-error.ts +58 -0
- package/wire/lib/strip-subagent-isolation.ts +284 -0
- package/wire/lib/thinking-signature.ts +156 -0
- package/wire/lib/token-ruler/bpe-core.ts +426 -0
- package/wire/lib/token-ruler/index.ts +2 -0
- package/wire/lib/token-ruler/ruler.ts +79 -0
- package/wire/lib/token-ruler/vocab/claude-ranks.ts +16 -0
- package/wire/lib/token-ruler/vocab/claude.ts +22 -0
- package/wire/lib/token-ruler/vocab/o200k-ranks.ts +18 -0
- package/wire/lib/token-ruler/vocab/o200k.ts +19 -0
- package/wire/lib/tool-name-map.ts +271 -0
- package/wire/lib/tool-schema.ts +353 -0
- package/wire/lib/usage-observer.ts +125 -0
- package/wire/providers/anthropic/adaptive-thinking.ts +291 -0
- package/wire/providers/anthropic/beta-headers.ts +165 -0
- package/wire/providers/anthropic/index.ts +10 -0
- package/wire/providers/anthropic/request.ts +607 -0
- package/wire/providers/anthropic/response.ts +155 -0
- package/wire/providers/anthropic/streaming.ts +315 -0
- package/wire/providers/anthropic/web-search-domains.ts +49 -0
- package/wire/providers/chatgpt/common.ts +136 -0
- package/wire/providers/chatgpt/index.ts +9 -0
- package/wire/providers/chatgpt/request.ts +959 -0
- package/wire/providers/chatgpt/streaming.ts +1007 -0
- package/wire/providers/chatgpt/web-search.ts +63 -0
- package/wire/providers/grok/web-search.ts +87 -0
- package/wire/providers/kimi/web-search.ts +145 -0
- package/wire/providers/upstream-deny.ts +355 -0
- package/wire/providers/upstream-request.ts +619 -0
|
@@ -0,0 +1,327 @@
|
|
|
1
|
+
import { Schema as S } from "effect";
|
|
2
|
+
|
|
3
|
+
export const ServiceTier = S.Literals(["auto", "default", "flex", "priority"]);
|
|
4
|
+
export type TServiceTier = S.Schema.Type<typeof ServiceTier>;
|
|
5
|
+
|
|
6
|
+
/** Positive token-count budget (user pin or catalog alternative). */
|
|
7
|
+
export const InputTokenLimit = S.Int.check(S.isGreaterThan(0));
|
|
8
|
+
export type TInputTokenLimit = S.Schema.Type<typeof InputTokenLimit>;
|
|
9
|
+
|
|
10
|
+
/**
|
|
11
|
+
* Reasoning-effort floor a user can pin per model. A request asking for less
|
|
12
|
+
* (or omitting effort) is raised to it; a request asking for more keeps its
|
|
13
|
+
* own. Exists because clients pick effort for the model they THINK they talk
|
|
14
|
+
* to (Claude Code sends `low` for an unrecognised alias), which can starve a
|
|
15
|
+
* model that quits early at low effort.
|
|
16
|
+
*/
|
|
17
|
+
export const MinReasoningEffort = S.Literals([
|
|
18
|
+
"low",
|
|
19
|
+
"medium",
|
|
20
|
+
"high",
|
|
21
|
+
"xhigh",
|
|
22
|
+
]);
|
|
23
|
+
export type TMinReasoningEffort = S.Schema.Type<typeof MinReasoningEffort>;
|
|
24
|
+
|
|
25
|
+
/** Ascending effort order used to compare a request against the floor. */
|
|
26
|
+
const EFFORT_RANK: Readonly<Record<string, number>> = {
|
|
27
|
+
none: 0,
|
|
28
|
+
minimal: 1,
|
|
29
|
+
low: 2,
|
|
30
|
+
medium: 3,
|
|
31
|
+
high: 4,
|
|
32
|
+
xhigh: 5,
|
|
33
|
+
max: 6,
|
|
34
|
+
};
|
|
35
|
+
|
|
36
|
+
/**
|
|
37
|
+
* Raise `effort` to `floor`. `undefined` effort (provider default) is raised
|
|
38
|
+
* too; an explicit `none` is the caller turning reasoning off and is kept.
|
|
39
|
+
*/
|
|
40
|
+
export const applyReasoningEffortFloor = <TEffort extends string>(
|
|
41
|
+
effort: TEffort | undefined,
|
|
42
|
+
floor: TMinReasoningEffort | undefined,
|
|
43
|
+
): TEffort | TMinReasoningEffort | undefined => {
|
|
44
|
+
if (floor === undefined || effort === "none") return effort;
|
|
45
|
+
if (effort === undefined) return floor;
|
|
46
|
+
return (EFFORT_RANK[effort] ?? 0) < (EFFORT_RANK[floor] ?? 0)
|
|
47
|
+
? floor
|
|
48
|
+
: effort;
|
|
49
|
+
};
|
|
50
|
+
|
|
51
|
+
/** Sparse user overrides. Add named, schema-checked settings here as they ship. */
|
|
52
|
+
export const ModelSettings = S.Struct({
|
|
53
|
+
service_tier: S.optional(ServiceTier),
|
|
54
|
+
/**
|
|
55
|
+
* Exact selected input budget. Absence means "Provider default" (latest
|
|
56
|
+
* provider report, else authored/inherited fallback) — never a snapshot
|
|
57
|
+
* written into config.
|
|
58
|
+
*/
|
|
59
|
+
input_token_limit: S.optional(InputTokenLimit),
|
|
60
|
+
min_reasoning_effort: S.optional(MinReasoningEffort),
|
|
61
|
+
});
|
|
62
|
+
export type TModelSettings = S.Schema.Type<typeof ModelSettings>;
|
|
63
|
+
|
|
64
|
+
export const ModelSettingsOverrides = S.Record(S.String, ModelSettings);
|
|
65
|
+
export type TModelSettingsOverrides = S.Schema.Type<
|
|
66
|
+
typeof ModelSettingsOverrides
|
|
67
|
+
>;
|
|
68
|
+
|
|
69
|
+
export const ServiceTierSetting = S.Struct({
|
|
70
|
+
label: S.String,
|
|
71
|
+
description: S.String,
|
|
72
|
+
options: S.Array(S.Struct({ value: ServiceTier, label: S.String })),
|
|
73
|
+
default: S.optional(ServiceTier),
|
|
74
|
+
/** Catalog opt-in: matched same-provider successor models may borrow this control. */
|
|
75
|
+
inherit_to_successors: S.optional(S.Boolean),
|
|
76
|
+
});
|
|
77
|
+
|
|
78
|
+
export const InputTokenLimitSetting = S.Struct({
|
|
79
|
+
label: S.String,
|
|
80
|
+
description: S.String,
|
|
81
|
+
options: S.Array(S.Struct({ value: InputTokenLimit, label: S.String })),
|
|
82
|
+
/**
|
|
83
|
+
* Latest provider-reported window for UI "Provider default" labelling
|
|
84
|
+
* only — never a pin option and never written into config.
|
|
85
|
+
*/
|
|
86
|
+
reported: S.optional(InputTokenLimit),
|
|
87
|
+
});
|
|
88
|
+
|
|
89
|
+
export const MinReasoningEffortSetting = S.Struct({
|
|
90
|
+
label: S.String,
|
|
91
|
+
description: S.String,
|
|
92
|
+
options: S.Array(S.Struct({ value: MinReasoningEffort, label: S.String })),
|
|
93
|
+
});
|
|
94
|
+
|
|
95
|
+
/** Catalog-owned controls and defaults, shared by discovery, UI and dispatch. */
|
|
96
|
+
export const ModelSettingDefinitions = S.Struct({
|
|
97
|
+
service_tier: S.optional(ServiceTierSetting),
|
|
98
|
+
input_token_limit: S.optional(InputTokenLimitSetting),
|
|
99
|
+
min_reasoning_effort: S.optional(MinReasoningEffortSetting),
|
|
100
|
+
});
|
|
101
|
+
export type TModelSettingDefinitions = S.Schema.Type<
|
|
102
|
+
typeof ModelSettingDefinitions
|
|
103
|
+
>;
|
|
104
|
+
|
|
105
|
+
const isSupportedServiceTier = (
|
|
106
|
+
setting: TModelSettingDefinitions["service_tier"],
|
|
107
|
+
value: TServiceTier | undefined,
|
|
108
|
+
): value is TServiceTier =>
|
|
109
|
+
value !== undefined &&
|
|
110
|
+
setting?.options.some((option) => option.value === value) === true;
|
|
111
|
+
|
|
112
|
+
const isSupportedInputTokenLimit = (
|
|
113
|
+
setting: TModelSettingDefinitions["input_token_limit"],
|
|
114
|
+
value: number | undefined,
|
|
115
|
+
): value is number =>
|
|
116
|
+
value !== undefined &&
|
|
117
|
+
setting?.options.some((option) => option.value === value) === true;
|
|
118
|
+
|
|
119
|
+
export const resolveModelSettings = (
|
|
120
|
+
definitions: TModelSettingDefinitions | undefined,
|
|
121
|
+
overrides: TModelSettings | undefined,
|
|
122
|
+
): TModelSettings => {
|
|
123
|
+
const resolved: {
|
|
124
|
+
-readonly [K in keyof TModelSettings]: TModelSettings[K];
|
|
125
|
+
} = {};
|
|
126
|
+
|
|
127
|
+
const tierSetting = definitions?.service_tier;
|
|
128
|
+
if (tierSetting !== undefined) {
|
|
129
|
+
const override = overrides?.service_tier;
|
|
130
|
+
const value = isSupportedServiceTier(tierSetting, override)
|
|
131
|
+
? override
|
|
132
|
+
: tierSetting.default;
|
|
133
|
+
if (isSupportedServiceTier(tierSetting, value)) {
|
|
134
|
+
resolved.service_tier = value;
|
|
135
|
+
}
|
|
136
|
+
}
|
|
137
|
+
|
|
138
|
+
const limitSetting = definitions?.input_token_limit;
|
|
139
|
+
if (limitSetting !== undefined) {
|
|
140
|
+
const override = overrides?.input_token_limit;
|
|
141
|
+
if (isSupportedInputTokenLimit(limitSetting, override)) {
|
|
142
|
+
resolved.input_token_limit = override;
|
|
143
|
+
}
|
|
144
|
+
}
|
|
145
|
+
|
|
146
|
+
const effortSetting = definitions?.min_reasoning_effort;
|
|
147
|
+
const effortOverride = overrides?.min_reasoning_effort;
|
|
148
|
+
if (
|
|
149
|
+
effortSetting !== undefined &&
|
|
150
|
+
effortOverride !== undefined &&
|
|
151
|
+
effortSetting.options.some((option) => option.value === effortOverride)
|
|
152
|
+
) {
|
|
153
|
+
resolved.min_reasoning_effort = effortOverride;
|
|
154
|
+
}
|
|
155
|
+
|
|
156
|
+
return resolved;
|
|
157
|
+
};
|
|
158
|
+
|
|
159
|
+
export const updateModelSettings = (
|
|
160
|
+
overrides: TModelSettingsOverrides | undefined,
|
|
161
|
+
modelId: string,
|
|
162
|
+
patch: {
|
|
163
|
+
readonly service_tier?: TServiceTier | undefined;
|
|
164
|
+
readonly input_token_limit?: number | undefined;
|
|
165
|
+
readonly clearServiceTier?: boolean;
|
|
166
|
+
readonly clearInputTokenLimit?: boolean;
|
|
167
|
+
readonly min_reasoning_effort?: TMinReasoningEffort | undefined;
|
|
168
|
+
readonly clearMinReasoningEffort?: boolean;
|
|
169
|
+
},
|
|
170
|
+
): TModelSettingsOverrides => {
|
|
171
|
+
const next = { ...overrides };
|
|
172
|
+
const previous = next[modelId] ?? {};
|
|
173
|
+
const updated: {
|
|
174
|
+
-readonly [K in keyof TModelSettings]: TModelSettings[K];
|
|
175
|
+
} = { ...previous };
|
|
176
|
+
|
|
177
|
+
if (patch.clearServiceTier === true) {
|
|
178
|
+
delete updated.service_tier;
|
|
179
|
+
} else if (patch.service_tier !== undefined) {
|
|
180
|
+
updated.service_tier = patch.service_tier;
|
|
181
|
+
}
|
|
182
|
+
|
|
183
|
+
if (patch.clearInputTokenLimit === true) {
|
|
184
|
+
delete updated.input_token_limit;
|
|
185
|
+
} else if (patch.input_token_limit !== undefined) {
|
|
186
|
+
updated.input_token_limit = patch.input_token_limit;
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
if (patch.clearMinReasoningEffort === true) {
|
|
190
|
+
delete updated.min_reasoning_effort;
|
|
191
|
+
} else if (patch.min_reasoning_effort !== undefined) {
|
|
192
|
+
updated.min_reasoning_effort = patch.min_reasoning_effort;
|
|
193
|
+
}
|
|
194
|
+
|
|
195
|
+
if (Object.keys(updated).length > 0) {
|
|
196
|
+
next[modelId] = updated;
|
|
197
|
+
} else {
|
|
198
|
+
delete next[modelId];
|
|
199
|
+
}
|
|
200
|
+
return next;
|
|
201
|
+
};
|
|
202
|
+
|
|
203
|
+
/** Codex config calls this "fast"; its request/app-server tier is "priority". */
|
|
204
|
+
export const codexServiceTier = (value: unknown): string | null => {
|
|
205
|
+
if (value === "fast" || value === "priority") return "priority";
|
|
206
|
+
if (value === "flex") return "flex";
|
|
207
|
+
return null;
|
|
208
|
+
};
|
|
209
|
+
|
|
210
|
+
/**
|
|
211
|
+
* Pure precedence for the gateway/native input budget.
|
|
212
|
+
*
|
|
213
|
+
* 1. exact valid user pin (supported by the model's current options)
|
|
214
|
+
* 2. latest provider report
|
|
215
|
+
* 3. authored/inherited fallback
|
|
216
|
+
*
|
|
217
|
+
* Never min/max a valid pin with discovery. Stale pins fall through.
|
|
218
|
+
*/
|
|
219
|
+
export const resolveEffectiveInputTokenLimit = (params: {
|
|
220
|
+
readonly userLimit?: number | undefined;
|
|
221
|
+
readonly reportedLimit?: number | null | undefined;
|
|
222
|
+
readonly fallbackLimit?: number | null | undefined;
|
|
223
|
+
readonly supportedLimits?: ReadonlyArray<number> | undefined;
|
|
224
|
+
}): number | null => {
|
|
225
|
+
const supported = params.supportedLimits;
|
|
226
|
+
if (
|
|
227
|
+
params.userLimit !== undefined &&
|
|
228
|
+
(supported === undefined || supported.includes(params.userLimit))
|
|
229
|
+
) {
|
|
230
|
+
return params.userLimit;
|
|
231
|
+
}
|
|
232
|
+
if (params.reportedLimit != null) return params.reportedLimit;
|
|
233
|
+
if (params.fallbackLimit != null) return params.fallbackLimit;
|
|
234
|
+
return null;
|
|
235
|
+
};
|
|
236
|
+
|
|
237
|
+
/** Deduplicate positive token counts while preserving first-seen order. */
|
|
238
|
+
export const dedupeTokenLimits = (
|
|
239
|
+
values: ReadonlyArray<number | null | undefined>,
|
|
240
|
+
): ReadonlyArray<number> => {
|
|
241
|
+
const seen = new Set<number>();
|
|
242
|
+
const out: number[] = [];
|
|
243
|
+
for (const value of values) {
|
|
244
|
+
if (value == null || !Number.isInteger(value) || value <= 0) continue;
|
|
245
|
+
if (seen.has(value)) continue;
|
|
246
|
+
seen.add(value);
|
|
247
|
+
out.push(value);
|
|
248
|
+
}
|
|
249
|
+
return out;
|
|
250
|
+
};
|
|
251
|
+
|
|
252
|
+
export const formatTokenLimitLabel = (value: number): string => {
|
|
253
|
+
if (value >= 1_000_000 && value % 1_000 === 0) {
|
|
254
|
+
return `${value / 1_000_000}M tokens`;
|
|
255
|
+
}
|
|
256
|
+
if (value >= 1_000 && value % 1_000 === 0) {
|
|
257
|
+
const thousands = value / 1_000;
|
|
258
|
+
return Number.isInteger(thousands)
|
|
259
|
+
? `${thousands}K tokens`
|
|
260
|
+
: `${value.toLocaleString("en-US")} tokens`;
|
|
261
|
+
}
|
|
262
|
+
return `${value.toLocaleString("en-US")} tokens`;
|
|
263
|
+
};
|
|
264
|
+
|
|
265
|
+
/**
|
|
266
|
+
* Synthesize the context-window control from authored alternatives only.
|
|
267
|
+
* Transient provider reports are observation for the Provider-default label,
|
|
268
|
+
* never pin options (a 275k report must not become a sticky invalid pin after
|
|
269
|
+
* the next report drifts). Definitions are never inherited as part of the
|
|
270
|
+
* `settings` bag (service tiers stay entitlement-gated).
|
|
271
|
+
*/
|
|
272
|
+
export const synthesizeInputTokenLimitSetting = (params: {
|
|
273
|
+
readonly alternativeInputLimits?: ReadonlyArray<number> | undefined;
|
|
274
|
+
readonly reportedLimit?: number | null | undefined;
|
|
275
|
+
}): TModelSettingDefinitions["input_token_limit"] | undefined => {
|
|
276
|
+
// Authored/inherited alternatives are what make the control exist. A bare
|
|
277
|
+
// provider report without alternatives is observation only — not a setting.
|
|
278
|
+
// Explicit `[]` is "own empty list" and yields no control (and blocks
|
|
279
|
+
// inheritance upstream).
|
|
280
|
+
if (
|
|
281
|
+
params.alternativeInputLimits === undefined ||
|
|
282
|
+
params.alternativeInputLimits.length === 0
|
|
283
|
+
) {
|
|
284
|
+
return undefined;
|
|
285
|
+
}
|
|
286
|
+
const options = dedupeTokenLimits([...params.alternativeInputLimits]).map(
|
|
287
|
+
(value) => ({
|
|
288
|
+
value,
|
|
289
|
+
label: formatTokenLimitLabel(value),
|
|
290
|
+
}),
|
|
291
|
+
);
|
|
292
|
+
if (options.length === 0) return undefined;
|
|
293
|
+
const reported =
|
|
294
|
+
params.reportedLimit != null &&
|
|
295
|
+
Number.isInteger(params.reportedLimit) &&
|
|
296
|
+
params.reportedLimit > 0
|
|
297
|
+
? params.reportedLimit
|
|
298
|
+
: undefined;
|
|
299
|
+
return {
|
|
300
|
+
label: "Context window",
|
|
301
|
+
description:
|
|
302
|
+
"Sets the gateway/native context budget and overflow threshold for this model. It is not a provider wire parameter or a capacity upgrade. Provider default follows the latest reported window.",
|
|
303
|
+
options,
|
|
304
|
+
...(reported !== undefined ? { reported } : {}),
|
|
305
|
+
};
|
|
306
|
+
};
|
|
307
|
+
|
|
308
|
+
const MIN_REASONING_EFFORT_SETTING: TModelSettingDefinitions["min_reasoning_effort"] =
|
|
309
|
+
{
|
|
310
|
+
label: "Minimum effort",
|
|
311
|
+
description:
|
|
312
|
+
"Default uses whatever effort the caller requested. Pick a level to raise lower requests to it; a request asking for more keeps its own, and one turning reasoning off is left alone.",
|
|
313
|
+
options: [
|
|
314
|
+
{ value: "low", label: "Low" },
|
|
315
|
+
{ value: "medium", label: "Medium" },
|
|
316
|
+
{ value: "high", label: "High" },
|
|
317
|
+
{ value: "xhigh", label: "Extra high" },
|
|
318
|
+
],
|
|
319
|
+
};
|
|
320
|
+
|
|
321
|
+
/** The effort-floor control exists on every reasoning-capable model. */
|
|
322
|
+
export const synthesizeMinReasoningEffortSetting = (
|
|
323
|
+
capabilities: ReadonlyArray<string> | undefined,
|
|
324
|
+
): TModelSettingDefinitions["min_reasoning_effort"] | undefined =>
|
|
325
|
+
capabilities?.includes("reasoning") === true
|
|
326
|
+
? MIN_REASONING_EFFORT_SETTING
|
|
327
|
+
: undefined;
|