@build-astron-co/nimbus 0.2.0
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/LICENSE +21 -0
- package/README.md +628 -0
- package/bin/nimbus +38 -0
- package/package.json +80 -0
- package/src/__tests__/app.test.ts +76 -0
- package/src/__tests__/audit.test.ts +877 -0
- package/src/__tests__/circuit-breaker.test.ts +116 -0
- package/src/__tests__/cli-run.test.ts +115 -0
- package/src/__tests__/context-manager.test.ts +502 -0
- package/src/__tests__/context.test.ts +242 -0
- package/src/__tests__/enterprise.test.ts +401 -0
- package/src/__tests__/generator.test.ts +433 -0
- package/src/__tests__/hooks.test.ts +582 -0
- package/src/__tests__/init.test.ts +436 -0
- package/src/__tests__/intent-parser.test.ts +229 -0
- package/src/__tests__/llm-router.test.ts +209 -0
- package/src/__tests__/lsp.test.ts +293 -0
- package/src/__tests__/modes.test.ts +336 -0
- package/src/__tests__/permissions.test.ts +338 -0
- package/src/__tests__/serve.test.ts +275 -0
- package/src/__tests__/sessions.test.ts +227 -0
- package/src/__tests__/sharing.test.ts +288 -0
- package/src/__tests__/snapshots.test.ts +581 -0
- package/src/__tests__/state-db.test.ts +334 -0
- package/src/__tests__/stream-with-tools.test.ts +732 -0
- package/src/__tests__/subagents.test.ts +176 -0
- package/src/__tests__/system-prompt.test.ts +169 -0
- package/src/__tests__/tool-converter.test.ts +256 -0
- package/src/__tests__/tool-schemas.test.ts +397 -0
- package/src/__tests__/tools.test.ts +143 -0
- package/src/__tests__/version.test.ts +49 -0
- package/src/agent/compaction-agent.ts +227 -0
- package/src/agent/context-manager.ts +435 -0
- package/src/agent/context.ts +427 -0
- package/src/agent/deploy-preview.ts +426 -0
- package/src/agent/index.ts +68 -0
- package/src/agent/loop.ts +717 -0
- package/src/agent/modes.ts +429 -0
- package/src/agent/permissions.ts +466 -0
- package/src/agent/subagents/base.ts +116 -0
- package/src/agent/subagents/cost.ts +51 -0
- package/src/agent/subagents/explore.ts +42 -0
- package/src/agent/subagents/general.ts +54 -0
- package/src/agent/subagents/index.ts +102 -0
- package/src/agent/subagents/infra.ts +59 -0
- package/src/agent/subagents/security.ts +69 -0
- package/src/agent/system-prompt.ts +436 -0
- package/src/app.ts +122 -0
- package/src/audit/activity-log.ts +290 -0
- package/src/audit/compliance-checker.ts +540 -0
- package/src/audit/cost-tracker.ts +318 -0
- package/src/audit/index.ts +23 -0
- package/src/audit/security-scanner.ts +596 -0
- package/src/auth/guard.ts +75 -0
- package/src/auth/index.ts +56 -0
- package/src/auth/oauth.ts +455 -0
- package/src/auth/providers.ts +470 -0
- package/src/auth/sso.ts +113 -0
- package/src/auth/store.ts +505 -0
- package/src/auth/types.ts +187 -0
- package/src/build.ts +141 -0
- package/src/cli/index.ts +16 -0
- package/src/cli/init.ts +854 -0
- package/src/cli/openapi-spec.ts +356 -0
- package/src/cli/run.ts +237 -0
- package/src/cli/serve-auth.ts +80 -0
- package/src/cli/serve.ts +462 -0
- package/src/cli/web.ts +67 -0
- package/src/cli.ts +1417 -0
- package/src/clients/core-engine-client.ts +227 -0
- package/src/clients/enterprise-client.ts +334 -0
- package/src/clients/generator-client.ts +351 -0
- package/src/clients/git-client.ts +627 -0
- package/src/clients/github-client.ts +410 -0
- package/src/clients/helm-client.ts +504 -0
- package/src/clients/index.ts +80 -0
- package/src/clients/k8s-client.ts +497 -0
- package/src/clients/llm-client.ts +161 -0
- package/src/clients/rest-client.ts +130 -0
- package/src/clients/service-discovery.ts +33 -0
- package/src/clients/terraform-client.ts +482 -0
- package/src/clients/tools-client.ts +1843 -0
- package/src/clients/ws-client.ts +115 -0
- package/src/commands/analyze/index.ts +352 -0
- package/src/commands/apply/helm.ts +473 -0
- package/src/commands/apply/index.ts +213 -0
- package/src/commands/apply/k8s.ts +454 -0
- package/src/commands/apply/terraform.ts +582 -0
- package/src/commands/ask.ts +167 -0
- package/src/commands/audit/index.ts +238 -0
- package/src/commands/auth-cloud.ts +294 -0
- package/src/commands/auth-list.ts +134 -0
- package/src/commands/auth-profile.ts +121 -0
- package/src/commands/auth-status.ts +141 -0
- package/src/commands/aws/ec2.ts +501 -0
- package/src/commands/aws/iam.ts +397 -0
- package/src/commands/aws/index.ts +133 -0
- package/src/commands/aws/lambda.ts +396 -0
- package/src/commands/aws/rds.ts +439 -0
- package/src/commands/aws/s3.ts +439 -0
- package/src/commands/aws/vpc.ts +393 -0
- package/src/commands/aws-discover.ts +649 -0
- package/src/commands/aws-terraform.ts +805 -0
- package/src/commands/azure/aks.ts +376 -0
- package/src/commands/azure/functions.ts +253 -0
- package/src/commands/azure/index.ts +116 -0
- package/src/commands/azure/storage.ts +478 -0
- package/src/commands/azure/vm.ts +355 -0
- package/src/commands/billing/index.ts +256 -0
- package/src/commands/chat.ts +314 -0
- package/src/commands/config.ts +346 -0
- package/src/commands/cost/cloud-cost-estimator.ts +266 -0
- package/src/commands/cost/estimator.ts +79 -0
- package/src/commands/cost/index.ts +594 -0
- package/src/commands/cost/parsers/terraform.ts +273 -0
- package/src/commands/cost/parsers/types.ts +25 -0
- package/src/commands/cost/pricing/aws.ts +544 -0
- package/src/commands/cost/pricing/azure.ts +499 -0
- package/src/commands/cost/pricing/gcp.ts +396 -0
- package/src/commands/cost/pricing/index.ts +40 -0
- package/src/commands/demo.ts +250 -0
- package/src/commands/doctor.ts +794 -0
- package/src/commands/drift/index.ts +439 -0
- package/src/commands/explain.ts +277 -0
- package/src/commands/feedback.ts +389 -0
- package/src/commands/fix.ts +324 -0
- package/src/commands/fs/index.ts +402 -0
- package/src/commands/gcp/compute.ts +325 -0
- package/src/commands/gcp/functions.ts +271 -0
- package/src/commands/gcp/gke.ts +438 -0
- package/src/commands/gcp/iam.ts +344 -0
- package/src/commands/gcp/index.ts +129 -0
- package/src/commands/gcp/storage.ts +284 -0
- package/src/commands/generate-helm.ts +1249 -0
- package/src/commands/generate-k8s.ts +1560 -0
- package/src/commands/generate-terraform.ts +1460 -0
- package/src/commands/gh/index.ts +863 -0
- package/src/commands/git/index.ts +1343 -0
- package/src/commands/helm/index.ts +1126 -0
- package/src/commands/help.ts +539 -0
- package/src/commands/history.ts +142 -0
- package/src/commands/import.ts +868 -0
- package/src/commands/index.ts +367 -0
- package/src/commands/init.ts +1046 -0
- package/src/commands/k8s/index.ts +1137 -0
- package/src/commands/login.ts +631 -0
- package/src/commands/logout.ts +83 -0
- package/src/commands/onboarding.ts +228 -0
- package/src/commands/plan/display.ts +279 -0
- package/src/commands/plan/index.ts +599 -0
- package/src/commands/preview.ts +452 -0
- package/src/commands/questionnaire.ts +1270 -0
- package/src/commands/resume.ts +55 -0
- package/src/commands/team/index.ts +346 -0
- package/src/commands/template.ts +232 -0
- package/src/commands/tf/index.ts +1034 -0
- package/src/commands/upgrade.ts +550 -0
- package/src/commands/usage/index.ts +134 -0
- package/src/commands/version.ts +170 -0
- package/src/compat/index.ts +2 -0
- package/src/compat/runtime.ts +12 -0
- package/src/compat/sqlite.ts +107 -0
- package/src/config/index.ts +17 -0
- package/src/config/manager.ts +530 -0
- package/src/config/safety-policy.ts +358 -0
- package/src/config/schema.ts +125 -0
- package/src/config/types.ts +527 -0
- package/src/context/context-db.ts +199 -0
- package/src/demo/index.ts +349 -0
- package/src/demo/scenarios/full-journey.ts +229 -0
- package/src/demo/scenarios/getting-started.ts +127 -0
- package/src/demo/scenarios/helm-release.ts +341 -0
- package/src/demo/scenarios/k8s-deployment.ts +194 -0
- package/src/demo/scenarios/terraform-vpc.ts +170 -0
- package/src/demo/types.ts +92 -0
- package/src/engine/cost-estimator.ts +438 -0
- package/src/engine/diagram-generator.ts +256 -0
- package/src/engine/drift-detector.ts +902 -0
- package/src/engine/executor.ts +1035 -0
- package/src/engine/index.ts +76 -0
- package/src/engine/orchestrator.ts +636 -0
- package/src/engine/planner.ts +720 -0
- package/src/engine/safety.ts +743 -0
- package/src/engine/verifier.ts +770 -0
- package/src/enterprise/audit.ts +348 -0
- package/src/enterprise/auth.ts +270 -0
- package/src/enterprise/billing.ts +822 -0
- package/src/enterprise/index.ts +17 -0
- package/src/enterprise/teams.ts +443 -0
- package/src/generator/best-practices.ts +1608 -0
- package/src/generator/helm.ts +630 -0
- package/src/generator/index.ts +37 -0
- package/src/generator/intent-parser.ts +514 -0
- package/src/generator/kubernetes.ts +976 -0
- package/src/generator/terraform.ts +1867 -0
- package/src/history/index.ts +8 -0
- package/src/history/manager.ts +322 -0
- package/src/history/types.ts +34 -0
- package/src/hooks/config.ts +432 -0
- package/src/hooks/engine.ts +391 -0
- package/src/hooks/index.ts +4 -0
- package/src/llm/auth-bridge.ts +198 -0
- package/src/llm/circuit-breaker.ts +140 -0
- package/src/llm/config-loader.ts +201 -0
- package/src/llm/cost-calculator.ts +171 -0
- package/src/llm/index.ts +8 -0
- package/src/llm/model-aliases.ts +115 -0
- package/src/llm/provider-registry.ts +63 -0
- package/src/llm/providers/anthropic.ts +433 -0
- package/src/llm/providers/bedrock.ts +477 -0
- package/src/llm/providers/google.ts +405 -0
- package/src/llm/providers/ollama.ts +767 -0
- package/src/llm/providers/openai-compatible.ts +340 -0
- package/src/llm/providers/openai.ts +328 -0
- package/src/llm/providers/openrouter.ts +338 -0
- package/src/llm/router.ts +1035 -0
- package/src/llm/types.ts +232 -0
- package/src/lsp/client.ts +298 -0
- package/src/lsp/languages.ts +116 -0
- package/src/lsp/manager.ts +278 -0
- package/src/mcp/client.ts +402 -0
- package/src/mcp/index.ts +5 -0
- package/src/mcp/manager.ts +133 -0
- package/src/nimbus.ts +214 -0
- package/src/plugins/index.ts +27 -0
- package/src/plugins/loader.ts +334 -0
- package/src/plugins/manager.ts +376 -0
- package/src/plugins/types.ts +284 -0
- package/src/scanners/cicd-scanner.ts +258 -0
- package/src/scanners/cloud-scanner.ts +466 -0
- package/src/scanners/framework-scanner.ts +469 -0
- package/src/scanners/iac-scanner.ts +388 -0
- package/src/scanners/index.ts +539 -0
- package/src/scanners/language-scanner.ts +276 -0
- package/src/scanners/package-manager-scanner.ts +277 -0
- package/src/scanners/types.ts +172 -0
- package/src/sessions/manager.ts +365 -0
- package/src/sessions/types.ts +44 -0
- package/src/sharing/sync.ts +296 -0
- package/src/sharing/viewer.ts +97 -0
- package/src/snapshots/index.ts +2 -0
- package/src/snapshots/manager.ts +530 -0
- package/src/state/artifacts.ts +147 -0
- package/src/state/audit.ts +137 -0
- package/src/state/billing.ts +240 -0
- package/src/state/checkpoints.ts +117 -0
- package/src/state/config.ts +67 -0
- package/src/state/conversations.ts +14 -0
- package/src/state/credentials.ts +154 -0
- package/src/state/db.ts +58 -0
- package/src/state/index.ts +26 -0
- package/src/state/messages.ts +115 -0
- package/src/state/projects.ts +123 -0
- package/src/state/schema.ts +236 -0
- package/src/state/sessions.ts +147 -0
- package/src/state/teams.ts +200 -0
- package/src/telemetry.ts +108 -0
- package/src/tools/aws-ops.ts +952 -0
- package/src/tools/azure-ops.ts +579 -0
- package/src/tools/file-ops.ts +593 -0
- package/src/tools/gcp-ops.ts +625 -0
- package/src/tools/git-ops.ts +773 -0
- package/src/tools/github-ops.ts +799 -0
- package/src/tools/helm-ops.ts +943 -0
- package/src/tools/index.ts +17 -0
- package/src/tools/k8s-ops.ts +819 -0
- package/src/tools/schemas/converter.ts +184 -0
- package/src/tools/schemas/devops.ts +612 -0
- package/src/tools/schemas/index.ts +73 -0
- package/src/tools/schemas/standard.ts +1144 -0
- package/src/tools/schemas/types.ts +705 -0
- package/src/tools/terraform-ops.ts +862 -0
- package/src/types/ambient.d.ts +193 -0
- package/src/types/config.ts +83 -0
- package/src/types/drift.ts +116 -0
- package/src/types/enterprise.ts +335 -0
- package/src/types/index.ts +20 -0
- package/src/types/plan.ts +44 -0
- package/src/types/request.ts +65 -0
- package/src/types/response.ts +54 -0
- package/src/types/service.ts +51 -0
- package/src/ui/App.tsx +997 -0
- package/src/ui/DeployPreview.tsx +169 -0
- package/src/ui/Header.tsx +68 -0
- package/src/ui/InputBox.tsx +350 -0
- package/src/ui/MessageList.tsx +585 -0
- package/src/ui/PermissionPrompt.tsx +151 -0
- package/src/ui/StatusBar.tsx +158 -0
- package/src/ui/ToolCallDisplay.tsx +409 -0
- package/src/ui/chat-ui.ts +853 -0
- package/src/ui/index.ts +33 -0
- package/src/ui/ink/index.ts +711 -0
- package/src/ui/streaming.ts +176 -0
- package/src/ui/types.ts +57 -0
- package/src/utils/analytics.ts +72 -0
- package/src/utils/cost-warning.ts +27 -0
- package/src/utils/env.ts +46 -0
- package/src/utils/errors.ts +69 -0
- package/src/utils/event-bus.ts +38 -0
- package/src/utils/index.ts +24 -0
- package/src/utils/logger.ts +171 -0
- package/src/utils/rate-limiter.ts +121 -0
- package/src/utils/service-auth.ts +49 -0
- package/src/utils/validation.ts +53 -0
- package/src/version.ts +4 -0
- package/src/watcher/index.ts +163 -0
- package/src/wizard/approval.ts +383 -0
- package/src/wizard/index.ts +25 -0
- package/src/wizard/prompts.ts +338 -0
- package/src/wizard/types.ts +171 -0
- package/src/wizard/ui.ts +556 -0
- package/src/wizard/wizard.ts +304 -0
- package/tsconfig.json +24 -0
|
@@ -0,0 +1,140 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Provider Circuit Breaker
|
|
3
|
+
*
|
|
4
|
+
* Prevents cascading failures by tracking consecutive errors per provider.
|
|
5
|
+
* When a provider fails too many times in a row, its circuit "opens" and
|
|
6
|
+
* requests are skipped until a cooldown period elapses. After cooldown the
|
|
7
|
+
* circuit enters HALF_OPEN, allowing a single probe request to determine
|
|
8
|
+
* whether the provider has recovered.
|
|
9
|
+
*
|
|
10
|
+
* States:
|
|
11
|
+
* CLOSED → normal operation (all requests pass through)
|
|
12
|
+
* OPEN → provider is failing; skip until cooldown expires
|
|
13
|
+
* HALF_OPEN → cooldown elapsed; allow one probe request
|
|
14
|
+
*/
|
|
15
|
+
|
|
16
|
+
export type CircuitState = 'CLOSED' | 'OPEN' | 'HALF_OPEN';
|
|
17
|
+
|
|
18
|
+
interface ProviderCircuit {
|
|
19
|
+
state: CircuitState;
|
|
20
|
+
failures: number;
|
|
21
|
+
lastFailure: number;
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
const DEFAULT_FAILURE_THRESHOLD = 5;
|
|
25
|
+
const DEFAULT_COOLDOWN_MS = 60_000; // 60 seconds
|
|
26
|
+
|
|
27
|
+
export class ProviderCircuitBreaker {
|
|
28
|
+
private circuits: Map<string, ProviderCircuit> = new Map();
|
|
29
|
+
private failureThreshold: number;
|
|
30
|
+
private cooldownMs: number;
|
|
31
|
+
|
|
32
|
+
constructor(opts?: { failureThreshold?: number; cooldownMs?: number }) {
|
|
33
|
+
this.failureThreshold = opts?.failureThreshold ?? DEFAULT_FAILURE_THRESHOLD;
|
|
34
|
+
this.cooldownMs = opts?.cooldownMs ?? DEFAULT_COOLDOWN_MS;
|
|
35
|
+
}
|
|
36
|
+
|
|
37
|
+
/**
|
|
38
|
+
* Check whether a provider is available for requests.
|
|
39
|
+
* Returns false only when the circuit is OPEN and cooldown hasn't elapsed.
|
|
40
|
+
*/
|
|
41
|
+
isAvailable(provider: string): boolean {
|
|
42
|
+
const circuit = this.circuits.get(provider);
|
|
43
|
+
if (!circuit) {
|
|
44
|
+
return true;
|
|
45
|
+
}
|
|
46
|
+
|
|
47
|
+
if (circuit.state === 'CLOSED') {
|
|
48
|
+
return true;
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
if (circuit.state === 'OPEN') {
|
|
52
|
+
const elapsed = Date.now() - circuit.lastFailure;
|
|
53
|
+
if (elapsed >= this.cooldownMs) {
|
|
54
|
+
// Transition to HALF_OPEN: allow a single probe
|
|
55
|
+
circuit.state = 'HALF_OPEN';
|
|
56
|
+
return true;
|
|
57
|
+
}
|
|
58
|
+
return false;
|
|
59
|
+
}
|
|
60
|
+
|
|
61
|
+
// HALF_OPEN: allow probe
|
|
62
|
+
return true;
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
/**
|
|
66
|
+
* Record a successful request. Resets the circuit to CLOSED.
|
|
67
|
+
*/
|
|
68
|
+
recordSuccess(provider: string): void {
|
|
69
|
+
const circuit = this.circuits.get(provider);
|
|
70
|
+
if (circuit) {
|
|
71
|
+
circuit.state = 'CLOSED';
|
|
72
|
+
circuit.failures = 0;
|
|
73
|
+
}
|
|
74
|
+
}
|
|
75
|
+
|
|
76
|
+
/**
|
|
77
|
+
* Record a failed request. Increments the failure counter and may
|
|
78
|
+
* open the circuit if the threshold is exceeded.
|
|
79
|
+
*/
|
|
80
|
+
recordFailure(provider: string): void {
|
|
81
|
+
let circuit = this.circuits.get(provider);
|
|
82
|
+
if (!circuit) {
|
|
83
|
+
circuit = { state: 'CLOSED', failures: 0, lastFailure: 0 };
|
|
84
|
+
this.circuits.set(provider, circuit);
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
circuit.failures++;
|
|
88
|
+
circuit.lastFailure = Date.now();
|
|
89
|
+
|
|
90
|
+
if (circuit.failures >= this.failureThreshold && circuit.state !== 'OPEN') {
|
|
91
|
+
circuit.state = 'OPEN';
|
|
92
|
+
// Emit a visible warning when a provider circuit opens
|
|
93
|
+
if (process.stderr.isTTY) {
|
|
94
|
+
process.stderr.write(
|
|
95
|
+
`\x1b[33m Warning: Provider '${provider}' disabled after ${this.failureThreshold} consecutive failures. Will retry in ${Math.round(this.cooldownMs / 1000)}s.\x1b[0m\n`
|
|
96
|
+
);
|
|
97
|
+
}
|
|
98
|
+
} else if (circuit.failures >= this.failureThreshold) {
|
|
99
|
+
circuit.state = 'OPEN';
|
|
100
|
+
}
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
/**
|
|
104
|
+
* Get the current state of a provider's circuit.
|
|
105
|
+
*/
|
|
106
|
+
getState(provider: string): CircuitState {
|
|
107
|
+
return this.circuits.get(provider)?.state ?? 'CLOSED';
|
|
108
|
+
}
|
|
109
|
+
|
|
110
|
+
/**
|
|
111
|
+
* Reset a specific provider's circuit (e.g., after manual recovery).
|
|
112
|
+
*/
|
|
113
|
+
reset(provider: string): void {
|
|
114
|
+
this.circuits.delete(provider);
|
|
115
|
+
}
|
|
116
|
+
|
|
117
|
+
/**
|
|
118
|
+
* Reset all circuits.
|
|
119
|
+
*/
|
|
120
|
+
resetAll(): void {
|
|
121
|
+
this.circuits.clear();
|
|
122
|
+
}
|
|
123
|
+
|
|
124
|
+
/**
|
|
125
|
+
* Get the names of all providers whose circuits are currently OPEN.
|
|
126
|
+
* Useful for surfacing circuit breaker state in the TUI.
|
|
127
|
+
*/
|
|
128
|
+
getOpenCircuits(): string[] {
|
|
129
|
+
const open: string[] = [];
|
|
130
|
+
for (const [name, circuit] of this.circuits) {
|
|
131
|
+
if (circuit.state === 'OPEN') {
|
|
132
|
+
const elapsed = Date.now() - circuit.lastFailure;
|
|
133
|
+
if (elapsed < this.cooldownMs) {
|
|
134
|
+
open.push(name);
|
|
135
|
+
}
|
|
136
|
+
}
|
|
137
|
+
}
|
|
138
|
+
return open;
|
|
139
|
+
}
|
|
140
|
+
}
|
|
@@ -0,0 +1,201 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* LLM Config Loader
|
|
3
|
+
*
|
|
4
|
+
* Reads ~/.nimbus/config.yaml and maps LLM-related settings
|
|
5
|
+
* to RouterConfig for the LLM Router.
|
|
6
|
+
*/
|
|
7
|
+
|
|
8
|
+
import * as fs from 'fs';
|
|
9
|
+
import * as path from 'path';
|
|
10
|
+
import * as os from 'os';
|
|
11
|
+
import { logger } from '../utils';
|
|
12
|
+
import type { RouterConfig } from './router';
|
|
13
|
+
|
|
14
|
+
/**
|
|
15
|
+
* Load LLM configuration from ~/.nimbus/config.yaml
|
|
16
|
+
* Returns a Partial<RouterConfig> that can be passed to LLMRouter constructor
|
|
17
|
+
*/
|
|
18
|
+
export function loadLLMConfig(): Partial<RouterConfig> {
|
|
19
|
+
const configPath =
|
|
20
|
+
process.env.NIMBUS_CONFIG_PATH || path.join(os.homedir(), '.nimbus', 'config.yaml');
|
|
21
|
+
|
|
22
|
+
if (!fs.existsSync(configPath)) {
|
|
23
|
+
logger.info(`No config file found at ${configPath}, using defaults`);
|
|
24
|
+
return {};
|
|
25
|
+
}
|
|
26
|
+
|
|
27
|
+
try {
|
|
28
|
+
const content = fs.readFileSync(configPath, 'utf-8');
|
|
29
|
+
const parsed = parseSimpleYaml(content);
|
|
30
|
+
const llmSection = parsed.llm || {};
|
|
31
|
+
|
|
32
|
+
const config: Partial<RouterConfig> = {};
|
|
33
|
+
|
|
34
|
+
// Map default provider
|
|
35
|
+
if (llmSection.default_provider) {
|
|
36
|
+
config.defaultProvider = llmSection.default_provider;
|
|
37
|
+
}
|
|
38
|
+
|
|
39
|
+
// Map default model
|
|
40
|
+
if (llmSection.defaultModel) {
|
|
41
|
+
config.defaultModel = llmSection.defaultModel;
|
|
42
|
+
}
|
|
43
|
+
|
|
44
|
+
// Map cost optimization settings
|
|
45
|
+
const costOpt = llmSection.cost_optimization;
|
|
46
|
+
if (costOpt) {
|
|
47
|
+
config.costOptimization = {
|
|
48
|
+
enabled: costOpt.enabled ?? false,
|
|
49
|
+
cheapModel: costOpt.cheap_model || '',
|
|
50
|
+
expensiveModel: costOpt.expensive_model || '',
|
|
51
|
+
cheapModelFor: costOpt.use_cheap_model_for || [],
|
|
52
|
+
expensiveModelFor: costOpt.use_expensive_model_for || [],
|
|
53
|
+
};
|
|
54
|
+
}
|
|
55
|
+
|
|
56
|
+
// Map fallback settings
|
|
57
|
+
const fallback = llmSection.fallback;
|
|
58
|
+
if (fallback) {
|
|
59
|
+
config.fallback = {
|
|
60
|
+
enabled: fallback.enabled ?? true,
|
|
61
|
+
providers: fallback.providers || [],
|
|
62
|
+
};
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
// Map token budget
|
|
66
|
+
if (llmSection.maxTokens) {
|
|
67
|
+
config.tokenBudget = {
|
|
68
|
+
maxTokensPerRequest: llmSection.maxTokens,
|
|
69
|
+
};
|
|
70
|
+
}
|
|
71
|
+
|
|
72
|
+
logger.info(`Loaded LLM config from ${configPath}`);
|
|
73
|
+
return config;
|
|
74
|
+
} catch (error: any) {
|
|
75
|
+
logger.warn(`Failed to load LLM config from ${configPath}: ${error.message}`);
|
|
76
|
+
return {};
|
|
77
|
+
}
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
/**
|
|
81
|
+
* Minimal YAML parser for flat/nested key-value configs.
|
|
82
|
+
* Handles the subset of YAML used by Nimbus config files.
|
|
83
|
+
*/
|
|
84
|
+
function parseSimpleYaml(content: string): Record<string, any> {
|
|
85
|
+
const result: Record<string, any> = {};
|
|
86
|
+
const lines = content.split('\n');
|
|
87
|
+
// Stack tracks: indent level, the parent object, and the last key set on that object
|
|
88
|
+
const stack: { indent: number; obj: Record<string, any>; lastKey?: string }[] = [
|
|
89
|
+
{ indent: -1, obj: result },
|
|
90
|
+
];
|
|
91
|
+
|
|
92
|
+
for (const rawLine of lines) {
|
|
93
|
+
// Skip comments and empty lines
|
|
94
|
+
const commentIdx = rawLine.indexOf('#');
|
|
95
|
+
const line = commentIdx >= 0 ? rawLine.slice(0, commentIdx) : rawLine;
|
|
96
|
+
if (line.trim() === '') {
|
|
97
|
+
continue;
|
|
98
|
+
}
|
|
99
|
+
|
|
100
|
+
const indent = line.length - line.trimStart().length;
|
|
101
|
+
const trimmed = line.trim();
|
|
102
|
+
|
|
103
|
+
// Handle list items (- value)
|
|
104
|
+
if (trimmed.startsWith('- ')) {
|
|
105
|
+
const listValue = trimmed.slice(2).trim();
|
|
106
|
+
// Pop stack to find the owner of the list key.
|
|
107
|
+
// If we're inside an empty object created by "key:" with no value,
|
|
108
|
+
// pop past it to find the frame that owns the key.
|
|
109
|
+
while (stack.length > 1 && stack[stack.length - 1].indent >= indent) {
|
|
110
|
+
stack.pop();
|
|
111
|
+
}
|
|
112
|
+
// If the current frame has no lastKey, it's a placeholder empty object --
|
|
113
|
+
// pop one more to reach the frame that holds the actual key reference.
|
|
114
|
+
let frame = stack[stack.length - 1];
|
|
115
|
+
if (!frame.lastKey && stack.length > 1) {
|
|
116
|
+
stack.pop();
|
|
117
|
+
frame = stack[stack.length - 1];
|
|
118
|
+
}
|
|
119
|
+
const parentObj = frame.obj;
|
|
120
|
+
const lastKey = frame.lastKey;
|
|
121
|
+
if (lastKey) {
|
|
122
|
+
if (!Array.isArray(parentObj[lastKey])) {
|
|
123
|
+
parentObj[lastKey] = [];
|
|
124
|
+
}
|
|
125
|
+
parentObj[lastKey].push(parseYamlValue(listValue));
|
|
126
|
+
}
|
|
127
|
+
continue;
|
|
128
|
+
}
|
|
129
|
+
|
|
130
|
+
// Handle key: value pairs
|
|
131
|
+
const colonIdx = trimmed.indexOf(':');
|
|
132
|
+
if (colonIdx === -1) {
|
|
133
|
+
continue;
|
|
134
|
+
}
|
|
135
|
+
|
|
136
|
+
const key = trimmed.slice(0, colonIdx).trim();
|
|
137
|
+
const rawValue = trimmed.slice(colonIdx + 1).trim();
|
|
138
|
+
|
|
139
|
+
// Pop stack to find parent at correct indent level
|
|
140
|
+
while (stack.length > 1 && stack[stack.length - 1].indent >= indent) {
|
|
141
|
+
stack.pop();
|
|
142
|
+
}
|
|
143
|
+
|
|
144
|
+
const frame = stack[stack.length - 1];
|
|
145
|
+
const current = frame.obj;
|
|
146
|
+
|
|
147
|
+
if (rawValue === '' || rawValue === undefined) {
|
|
148
|
+
// Nested object (or potentially a list -- will be converted if list items follow)
|
|
149
|
+
current[key] = {};
|
|
150
|
+
frame.lastKey = key;
|
|
151
|
+
stack.push({ indent, obj: current[key] });
|
|
152
|
+
} else {
|
|
153
|
+
current[key] = parseYamlValue(rawValue);
|
|
154
|
+
frame.lastKey = key;
|
|
155
|
+
}
|
|
156
|
+
}
|
|
157
|
+
|
|
158
|
+
return result;
|
|
159
|
+
}
|
|
160
|
+
|
|
161
|
+
/**
|
|
162
|
+
* Parse a YAML scalar value
|
|
163
|
+
*/
|
|
164
|
+
function parseYamlValue(value: string): any {
|
|
165
|
+
// Remove surrounding quotes
|
|
166
|
+
if (
|
|
167
|
+
(value.startsWith('"') && value.endsWith('"')) ||
|
|
168
|
+
(value.startsWith("'") && value.endsWith("'"))
|
|
169
|
+
) {
|
|
170
|
+
return value.slice(1, -1);
|
|
171
|
+
}
|
|
172
|
+
|
|
173
|
+
// Boolean
|
|
174
|
+
if (value === 'true') {
|
|
175
|
+
return true;
|
|
176
|
+
}
|
|
177
|
+
if (value === 'false') {
|
|
178
|
+
return false;
|
|
179
|
+
}
|
|
180
|
+
|
|
181
|
+
// Null
|
|
182
|
+
if (value === 'null' || value === '~') {
|
|
183
|
+
return null;
|
|
184
|
+
}
|
|
185
|
+
|
|
186
|
+
// Number
|
|
187
|
+
const num = Number(value);
|
|
188
|
+
if (!isNaN(num) && value !== '') {
|
|
189
|
+
return num;
|
|
190
|
+
}
|
|
191
|
+
|
|
192
|
+
// Inline list [a, b, c]
|
|
193
|
+
if (value.startsWith('[') && value.endsWith(']')) {
|
|
194
|
+
return value
|
|
195
|
+
.slice(1, -1)
|
|
196
|
+
.split(',')
|
|
197
|
+
.map(v => parseYamlValue(v.trim()));
|
|
198
|
+
}
|
|
199
|
+
|
|
200
|
+
return value;
|
|
201
|
+
}
|
|
@@ -0,0 +1,171 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Per-request cost calculation for LLM providers
|
|
3
|
+
*
|
|
4
|
+
* Pricing data is based on published rates from each provider.
|
|
5
|
+
* Prices are per 1K tokens (input / output).
|
|
6
|
+
* Ollama models are local and free.
|
|
7
|
+
* Unknown models return 0 cost with a warning log.
|
|
8
|
+
*/
|
|
9
|
+
|
|
10
|
+
import { logger } from '../utils';
|
|
11
|
+
|
|
12
|
+
export interface CostBreakdown {
|
|
13
|
+
/** Input token cost in USD */
|
|
14
|
+
input: number;
|
|
15
|
+
/** Output token cost in USD */
|
|
16
|
+
output: number;
|
|
17
|
+
}
|
|
18
|
+
|
|
19
|
+
export interface CostResult {
|
|
20
|
+
/** Total cost in USD */
|
|
21
|
+
costUSD: number;
|
|
22
|
+
/** Per-component breakdown */
|
|
23
|
+
breakdown: CostBreakdown;
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
/**
|
|
27
|
+
* Pricing entry: [inputPricePer1K, outputPricePer1K] in USD
|
|
28
|
+
*/
|
|
29
|
+
type PricingEntry = [input: number, output: number];
|
|
30
|
+
|
|
31
|
+
/**
|
|
32
|
+
* Pricing data by provider and model.
|
|
33
|
+
* Prices are in USD per 1,000 tokens.
|
|
34
|
+
*/
|
|
35
|
+
const PRICING: Record<string, Record<string, PricingEntry>> = {
|
|
36
|
+
anthropic: {
|
|
37
|
+
// Claude 4 family
|
|
38
|
+
'claude-4-opus': [0.015, 0.075],
|
|
39
|
+
'claude-opus-4-20250514': [0.015, 0.075],
|
|
40
|
+
'claude-4-sonnet': [0.003, 0.015],
|
|
41
|
+
'claude-sonnet-4-20250514': [0.003, 0.015],
|
|
42
|
+
'claude-4-haiku': [0.0008, 0.004],
|
|
43
|
+
'claude-haiku-4-20250514': [0.0008, 0.004],
|
|
44
|
+
// Claude 3.5 family
|
|
45
|
+
'claude-3.5-sonnet': [0.003, 0.015],
|
|
46
|
+
'claude-3-5-sonnet-20241022': [0.003, 0.015],
|
|
47
|
+
'claude-3.5-haiku': [0.0008, 0.004],
|
|
48
|
+
'claude-3-5-haiku-20241022': [0.0008, 0.004],
|
|
49
|
+
// Claude 3 family
|
|
50
|
+
'claude-3-opus': [0.015, 0.075],
|
|
51
|
+
'claude-3-opus-20240229': [0.015, 0.075],
|
|
52
|
+
'claude-3-haiku': [0.00025, 0.00125],
|
|
53
|
+
'claude-3-haiku-20240307': [0.00025, 0.00125],
|
|
54
|
+
},
|
|
55
|
+
openai: {
|
|
56
|
+
'gpt-4': [0.03, 0.06],
|
|
57
|
+
'gpt-4-turbo': [0.01, 0.03],
|
|
58
|
+
'gpt-4-turbo-2024-04-09': [0.01, 0.03],
|
|
59
|
+
'gpt-4o': [0.005, 0.015],
|
|
60
|
+
'gpt-4o-2024-11-20': [0.005, 0.015],
|
|
61
|
+
'gpt-4o-mini': [0.00015, 0.0006],
|
|
62
|
+
'gpt-4o-mini-2024-07-18': [0.00015, 0.0006],
|
|
63
|
+
'gpt-3.5-turbo': [0.0005, 0.0015],
|
|
64
|
+
},
|
|
65
|
+
google: {
|
|
66
|
+
'gemini-pro': [0.00025, 0.0005],
|
|
67
|
+
'gemini-1.5-pro': [0.00125, 0.005],
|
|
68
|
+
'gemini-1.5-flash': [0.000075, 0.0003],
|
|
69
|
+
'gemini-2.0-flash-exp': [0.000075, 0.0003],
|
|
70
|
+
},
|
|
71
|
+
ollama: {
|
|
72
|
+
// All local models are free
|
|
73
|
+
},
|
|
74
|
+
groq: {
|
|
75
|
+
'llama-3.1-70b-versatile': [0.00059, 0.00079],
|
|
76
|
+
'llama-3.1-8b-instant': [0.00005, 0.00008],
|
|
77
|
+
'llama-3.3-70b-versatile': [0.00059, 0.00079],
|
|
78
|
+
'mixtral-8x7b-32768': [0.00024, 0.00024],
|
|
79
|
+
'gemma2-9b-it': [0.0002, 0.0002],
|
|
80
|
+
},
|
|
81
|
+
together: {
|
|
82
|
+
'meta-llama/Llama-3.1-70B-Instruct-Turbo': [0.00088, 0.00088],
|
|
83
|
+
'meta-llama/Llama-3.1-8B-Instruct-Turbo': [0.00018, 0.00018],
|
|
84
|
+
'meta-llama/Llama-3.3-70B-Instruct-Turbo': [0.00088, 0.00088],
|
|
85
|
+
'mistralai/Mixtral-8x22B-Instruct-v0.1': [0.0012, 0.0012],
|
|
86
|
+
'Qwen/Qwen2.5-72B-Instruct-Turbo': [0.0012, 0.0012],
|
|
87
|
+
},
|
|
88
|
+
deepseek: {
|
|
89
|
+
'deepseek-chat': [0.00014, 0.00028],
|
|
90
|
+
'deepseek-coder': [0.00014, 0.00028],
|
|
91
|
+
'deepseek-reasoner': [0.00055, 0.00219],
|
|
92
|
+
},
|
|
93
|
+
fireworks: {
|
|
94
|
+
'accounts/fireworks/models/llama-v3p1-70b-instruct': [0.0009, 0.0009],
|
|
95
|
+
'accounts/fireworks/models/llama-v3p1-8b-instruct': [0.0002, 0.0002],
|
|
96
|
+
'accounts/fireworks/models/mixtral-8x22b-instruct': [0.0009, 0.0009],
|
|
97
|
+
'accounts/fireworks/models/qwen2p5-72b-instruct': [0.0009, 0.0009],
|
|
98
|
+
},
|
|
99
|
+
perplexity: {
|
|
100
|
+
'llama-3.1-sonar-large-128k-online': [0.001, 0.001],
|
|
101
|
+
'llama-3.1-sonar-small-128k-online': [0.0002, 0.0002],
|
|
102
|
+
'llama-3.1-sonar-huge-128k-online': [0.005, 0.005],
|
|
103
|
+
},
|
|
104
|
+
openrouter: {
|
|
105
|
+
// OpenRouter models use provider/model format; prices approximate the underlying provider
|
|
106
|
+
'anthropic/claude-sonnet-4-20250514': [0.003, 0.015],
|
|
107
|
+
'anthropic/claude-haiku-4-20250514': [0.0008, 0.004],
|
|
108
|
+
'anthropic/claude-opus-4-20250514': [0.015, 0.075],
|
|
109
|
+
'anthropic/claude-3.5-sonnet': [0.003, 0.015],
|
|
110
|
+
'openai/gpt-4o': [0.005, 0.015],
|
|
111
|
+
'openai/gpt-4o-mini': [0.00015, 0.0006],
|
|
112
|
+
'google/gemini-2.0-flash-exp': [0.000075, 0.0003],
|
|
113
|
+
'meta-llama/llama-3.1-405b-instruct': [0.003, 0.003],
|
|
114
|
+
'meta-llama/llama-3.1-70b-instruct': [0.0008, 0.0008],
|
|
115
|
+
'mistralai/mixtral-8x22b-instruct': [0.0009, 0.0009],
|
|
116
|
+
},
|
|
117
|
+
};
|
|
118
|
+
|
|
119
|
+
/**
|
|
120
|
+
* Calculate the cost for a single LLM request.
|
|
121
|
+
*
|
|
122
|
+
* @param provider - Provider name (e.g. "anthropic", "openai", "google", "ollama", "openrouter")
|
|
123
|
+
* @param model - Model identifier (e.g. "claude-sonnet-4-20250514", "gpt-4o")
|
|
124
|
+
* @param inputTokens - Number of input/prompt tokens
|
|
125
|
+
* @param outputTokens - Number of output/completion tokens
|
|
126
|
+
* @returns Cost result with total USD cost and per-component breakdown
|
|
127
|
+
*/
|
|
128
|
+
export function calculateCost(
|
|
129
|
+
provider: string,
|
|
130
|
+
model: string,
|
|
131
|
+
inputTokens: number,
|
|
132
|
+
outputTokens: number
|
|
133
|
+
): CostResult {
|
|
134
|
+
// Ollama is always free
|
|
135
|
+
if (provider === 'ollama') {
|
|
136
|
+
return { costUSD: 0, breakdown: { input: 0, output: 0 } };
|
|
137
|
+
}
|
|
138
|
+
|
|
139
|
+
const providerPricing = PRICING[provider];
|
|
140
|
+
if (!providerPricing) {
|
|
141
|
+
logger.warn(`No pricing data for provider "${provider}", returning zero cost`);
|
|
142
|
+
return { costUSD: 0, breakdown: { input: 0, output: 0 } };
|
|
143
|
+
}
|
|
144
|
+
|
|
145
|
+
const entry = providerPricing[model];
|
|
146
|
+
if (!entry) {
|
|
147
|
+
logger.warn(
|
|
148
|
+
`No pricing data for model "${model}" on provider "${provider}", returning zero cost`
|
|
149
|
+
);
|
|
150
|
+
return { costUSD: 0, breakdown: { input: 0, output: 0 } };
|
|
151
|
+
}
|
|
152
|
+
|
|
153
|
+
const [inputPricePer1K, outputPricePer1K] = entry;
|
|
154
|
+
const inputCost = (inputTokens / 1000) * inputPricePer1K;
|
|
155
|
+
const outputCost = (outputTokens / 1000) * outputPricePer1K;
|
|
156
|
+
|
|
157
|
+
return {
|
|
158
|
+
costUSD: inputCost + outputCost,
|
|
159
|
+
breakdown: {
|
|
160
|
+
input: inputCost,
|
|
161
|
+
output: outputCost,
|
|
162
|
+
},
|
|
163
|
+
};
|
|
164
|
+
}
|
|
165
|
+
|
|
166
|
+
/**
|
|
167
|
+
* Get available pricing data (useful for debugging / admin endpoints).
|
|
168
|
+
*/
|
|
169
|
+
export function getPricingData(): Record<string, Record<string, PricingEntry>> {
|
|
170
|
+
return PRICING;
|
|
171
|
+
}
|
package/src/llm/index.ts
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
export { LLMRouter } from './router';
|
|
2
|
+
export type { RouterConfig, ProviderInfo, StreamFallbackMeta } from './router';
|
|
3
|
+
export * from './types';
|
|
4
|
+
export { calculateCost } from './cost-calculator';
|
|
5
|
+
export { loadLLMConfig } from './config-loader';
|
|
6
|
+
export { resolveModelAlias, stripProviderPrefix } from './model-aliases';
|
|
7
|
+
export { detectProvider } from './provider-registry';
|
|
8
|
+
export * from './auth-bridge';
|
|
@@ -0,0 +1,115 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Model alias resolution
|
|
3
|
+
* Maps short names to full model identifiers
|
|
4
|
+
*/
|
|
5
|
+
|
|
6
|
+
const MODEL_ALIASES: Record<string, string> = {
|
|
7
|
+
// Anthropic shortcuts
|
|
8
|
+
sonnet: 'claude-sonnet-4-20250514',
|
|
9
|
+
haiku: 'claude-haiku-4-20250514',
|
|
10
|
+
opus: 'claude-opus-4-20250514',
|
|
11
|
+
claude: 'claude-sonnet-4-20250514',
|
|
12
|
+
|
|
13
|
+
// OpenAI shortcuts
|
|
14
|
+
gpt4: 'gpt-4o',
|
|
15
|
+
gpt4o: 'gpt-4o',
|
|
16
|
+
gpt4mini: 'gpt-4o-mini',
|
|
17
|
+
'4o': 'gpt-4o',
|
|
18
|
+
'4o-mini': 'gpt-4o-mini',
|
|
19
|
+
|
|
20
|
+
// Google shortcuts
|
|
21
|
+
gemini: 'gemini-2.0-flash-exp',
|
|
22
|
+
'gemini-flash': 'gemini-2.0-flash-exp',
|
|
23
|
+
'gemini-pro': 'gemini-1.5-pro',
|
|
24
|
+
|
|
25
|
+
// Groq shortcuts
|
|
26
|
+
groq: 'llama-3.1-70b-versatile',
|
|
27
|
+
'groq-fast': 'llama-3.1-8b-instant',
|
|
28
|
+
|
|
29
|
+
// Together shortcuts
|
|
30
|
+
together: 'meta-llama/Llama-3.1-70B-Instruct-Turbo',
|
|
31
|
+
'together-llama': 'meta-llama/Llama-3.1-70B-Instruct-Turbo',
|
|
32
|
+
|
|
33
|
+
// DeepSeek shortcuts
|
|
34
|
+
deepseek: 'deepseek-chat',
|
|
35
|
+
'deepseek-coder': 'deepseek-coder',
|
|
36
|
+
'deepseek-r1': 'deepseek-reasoner',
|
|
37
|
+
|
|
38
|
+
// Fireworks shortcuts
|
|
39
|
+
fireworks: 'accounts/fireworks/models/llama-v3p1-70b-instruct',
|
|
40
|
+
|
|
41
|
+
// Perplexity shortcuts
|
|
42
|
+
perplexity: 'llama-3.1-sonar-large-128k-online',
|
|
43
|
+
pplx: 'llama-3.1-sonar-large-128k-online',
|
|
44
|
+
sonar: 'llama-3.1-sonar-large-128k-online',
|
|
45
|
+
|
|
46
|
+
// Local shortcuts
|
|
47
|
+
llama: 'llama3.2',
|
|
48
|
+
mistral: 'mistral',
|
|
49
|
+
codellama: 'codellama',
|
|
50
|
+
};
|
|
51
|
+
|
|
52
|
+
/**
|
|
53
|
+
* Resolve a model alias to its full model identifier.
|
|
54
|
+
* Returns the original string if no alias match is found.
|
|
55
|
+
*
|
|
56
|
+
* @param modelOrAlias - Short alias or full model identifier
|
|
57
|
+
* @returns Resolved full model identifier
|
|
58
|
+
*/
|
|
59
|
+
export function resolveModelAlias(modelOrAlias: string): string {
|
|
60
|
+
return MODEL_ALIASES[modelOrAlias.toLowerCase()] || modelOrAlias;
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
/**
|
|
64
|
+
* Strip a "provider/" prefix from a model string.
|
|
65
|
+
*
|
|
66
|
+
* Model strings can be specified as "provider/model" (e.g., "anthropic/claude-sonnet-4-20250514").
|
|
67
|
+
* The provider prefix is used for routing but must be stripped before sending to the
|
|
68
|
+
* provider's API, which expects just the model ID (e.g., "claude-sonnet-4-20250514").
|
|
69
|
+
*
|
|
70
|
+
* Preserves multi-segment model IDs used by OpenRouter (e.g., "meta-llama/llama-3.1-405b")
|
|
71
|
+
* and Fireworks (e.g., "accounts/fireworks/models/llama-v3p3-70b-instruct") by only
|
|
72
|
+
* stripping prefixes that match known provider names.
|
|
73
|
+
*
|
|
74
|
+
* @param model - Model string, optionally with provider prefix
|
|
75
|
+
* @returns Model string with provider prefix removed
|
|
76
|
+
*/
|
|
77
|
+
export function stripProviderPrefix(model: string): string {
|
|
78
|
+
if (!model.includes('/')) {
|
|
79
|
+
return model;
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
const slashIndex = model.indexOf('/');
|
|
83
|
+
const prefix = model.slice(0, slashIndex).toLowerCase();
|
|
84
|
+
|
|
85
|
+
// Only strip if the prefix is a known provider name.
|
|
86
|
+
// This preserves OpenRouter-style "meta-llama/llama-3.1-405b" and
|
|
87
|
+
// Fireworks-style "accounts/fireworks/models/..." paths.
|
|
88
|
+
const KNOWN_PROVIDERS = new Set([
|
|
89
|
+
'anthropic',
|
|
90
|
+
'openai',
|
|
91
|
+
'google',
|
|
92
|
+
'ollama',
|
|
93
|
+
'groq',
|
|
94
|
+
'together',
|
|
95
|
+
'deepseek',
|
|
96
|
+
'fireworks',
|
|
97
|
+
'perplexity',
|
|
98
|
+
'bedrock',
|
|
99
|
+
]);
|
|
100
|
+
|
|
101
|
+
if (KNOWN_PROVIDERS.has(prefix)) {
|
|
102
|
+
return model.slice(slashIndex + 1);
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
return model;
|
|
106
|
+
}
|
|
107
|
+
|
|
108
|
+
/**
|
|
109
|
+
* Get a copy of all registered model aliases.
|
|
110
|
+
*
|
|
111
|
+
* @returns Record mapping alias names to full model identifiers
|
|
112
|
+
*/
|
|
113
|
+
export function getAliases(): Record<string, string> {
|
|
114
|
+
return { ...MODEL_ALIASES };
|
|
115
|
+
}
|
|
@@ -0,0 +1,63 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Provider Registry
|
|
3
|
+
* Auto-detects the provider from a model string
|
|
4
|
+
*/
|
|
5
|
+
|
|
6
|
+
/**
|
|
7
|
+
* Map of known provider prefixes used in "provider/model" format
|
|
8
|
+
*/
|
|
9
|
+
const PROVIDER_MAP: Record<string, string> = {
|
|
10
|
+
anthropic: 'anthropic',
|
|
11
|
+
openai: 'openai',
|
|
12
|
+
google: 'google',
|
|
13
|
+
'meta-llama': 'openrouter',
|
|
14
|
+
mistralai: 'openrouter',
|
|
15
|
+
groq: 'groq',
|
|
16
|
+
together: 'together',
|
|
17
|
+
deepseek: 'deepseek',
|
|
18
|
+
fireworks: 'fireworks',
|
|
19
|
+
perplexity: 'perplexity',
|
|
20
|
+
};
|
|
21
|
+
|
|
22
|
+
/**
|
|
23
|
+
* Detect the provider for a given model string.
|
|
24
|
+
*
|
|
25
|
+
* Detection strategy:
|
|
26
|
+
* 1. Explicit provider prefix (e.g., "groq/llama-3.1-70b") -- split on "/" and look up prefix
|
|
27
|
+
* 2. Model name pattern matching (e.g., "claude-*" -> anthropic, "gpt-*" -> openai)
|
|
28
|
+
* 3. Default to "anthropic" if no pattern matches
|
|
29
|
+
*
|
|
30
|
+
* @param model - Model identifier, optionally prefixed with "provider/"
|
|
31
|
+
* @returns Detected provider name
|
|
32
|
+
*/
|
|
33
|
+
export function detectProvider(model: string): string {
|
|
34
|
+
// Explicit provider prefix (e.g., "groq/llama-3.1-70b")
|
|
35
|
+
if (model.includes('/')) {
|
|
36
|
+
const prefix = model.split('/')[0].toLowerCase();
|
|
37
|
+
return PROVIDER_MAP[prefix] || 'openrouter';
|
|
38
|
+
}
|
|
39
|
+
|
|
40
|
+
// Model name pattern matching
|
|
41
|
+
if (model.startsWith('claude')) {
|
|
42
|
+
return 'anthropic';
|
|
43
|
+
}
|
|
44
|
+
if (model.startsWith('gpt')) {
|
|
45
|
+
return 'openai';
|
|
46
|
+
}
|
|
47
|
+
if (model.startsWith('gemini')) {
|
|
48
|
+
return 'google';
|
|
49
|
+
}
|
|
50
|
+
if (
|
|
51
|
+
model.startsWith('llama') ||
|
|
52
|
+
model.startsWith('mistral') ||
|
|
53
|
+
model.startsWith('codellama') ||
|
|
54
|
+
model.startsWith('phi')
|
|
55
|
+
) {
|
|
56
|
+
return 'ollama';
|
|
57
|
+
}
|
|
58
|
+
if (model.startsWith('deepseek')) {
|
|
59
|
+
return 'deepseek';
|
|
60
|
+
}
|
|
61
|
+
|
|
62
|
+
return 'anthropic'; // default
|
|
63
|
+
}
|