exovault-mcp-server 1.0.0
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/dist/auth.d.ts +41 -0
- package/dist/auth.d.ts.map +1 -0
- package/dist/auth.js +236 -0
- package/dist/auth.js.map +1 -0
- package/dist/auto-session.d.ts +39 -0
- package/dist/auto-session.d.ts.map +1 -0
- package/dist/auto-session.js +128 -0
- package/dist/auto-session.js.map +1 -0
- package/dist/buffer-persistence.d.ts +35 -0
- package/dist/buffer-persistence.d.ts.map +1 -0
- package/dist/buffer-persistence.js +110 -0
- package/dist/buffer-persistence.js.map +1 -0
- package/dist/coerce-params.d.ts +36 -0
- package/dist/coerce-params.d.ts.map +1 -0
- package/dist/coerce-params.js +120 -0
- package/dist/coerce-params.js.map +1 -0
- package/dist/crypto.d.ts +39 -0
- package/dist/crypto.d.ts.map +1 -0
- package/dist/crypto.js +119 -0
- package/dist/crypto.js.map +1 -0
- package/dist/db.d.ts +350 -0
- package/dist/db.d.ts.map +1 -0
- package/dist/db.js +866 -0
- package/dist/db.js.map +1 -0
- package/dist/embedding-config.d.ts +11 -0
- package/dist/embedding-config.d.ts.map +1 -0
- package/dist/embedding-config.js +24 -0
- package/dist/embedding-config.js.map +1 -0
- package/dist/entity-extraction.d.ts +22 -0
- package/dist/entity-extraction.d.ts.map +1 -0
- package/dist/entity-extraction.js +140 -0
- package/dist/entity-extraction.js.map +1 -0
- package/dist/episodic-headline.d.ts +6 -0
- package/dist/episodic-headline.d.ts.map +1 -0
- package/dist/episodic-headline.js +62 -0
- package/dist/episodic-headline.js.map +1 -0
- package/dist/error-sanitizer.d.ts +20 -0
- package/dist/error-sanitizer.d.ts.map +1 -0
- package/dist/error-sanitizer.js +54 -0
- package/dist/error-sanitizer.js.map +1 -0
- package/dist/extraction-budget.d.ts +39 -0
- package/dist/extraction-budget.d.ts.map +1 -0
- package/dist/extraction-budget.js +122 -0
- package/dist/extraction-budget.js.map +1 -0
- package/dist/extraction-llm.d.ts +22 -0
- package/dist/extraction-llm.d.ts.map +1 -0
- package/dist/extraction-llm.js +32 -0
- package/dist/extraction-llm.js.map +1 -0
- package/dist/extraction-prompt.d.ts +40 -0
- package/dist/extraction-prompt.d.ts.map +1 -0
- package/dist/extraction-prompt.js +176 -0
- package/dist/extraction-prompt.js.map +1 -0
- package/dist/gateway-client.d.ts +303 -0
- package/dist/gateway-client.d.ts.map +1 -0
- package/dist/gateway-client.js +285 -0
- package/dist/gateway-client.js.map +1 -0
- package/dist/gateway-init.d.ts +32 -0
- package/dist/gateway-init.d.ts.map +1 -0
- package/dist/gateway-init.js +71 -0
- package/dist/gateway-init.js.map +1 -0
- package/dist/index.d.ts +2 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +1242 -0
- package/dist/index.js.map +1 -0
- package/dist/infer-task-status.d.ts +7 -0
- package/dist/infer-task-status.d.ts.map +1 -0
- package/dist/infer-task-status.js +23 -0
- package/dist/infer-task-status.js.map +1 -0
- package/dist/normalize-agent-id.d.ts +21 -0
- package/dist/normalize-agent-id.d.ts.map +1 -0
- package/dist/normalize-agent-id.js +54 -0
- package/dist/normalize-agent-id.js.map +1 -0
- package/dist/openai.d.ts +14 -0
- package/dist/openai.d.ts.map +1 -0
- package/dist/openai.js +43 -0
- package/dist/openai.js.map +1 -0
- package/dist/rlm/actions.d.ts +31 -0
- package/dist/rlm/actions.d.ts.map +1 -0
- package/dist/rlm/actions.js +241 -0
- package/dist/rlm/actions.js.map +1 -0
- package/dist/rlm/benchmark.d.ts +2 -0
- package/dist/rlm/benchmark.d.ts.map +1 -0
- package/dist/rlm/benchmark.js +215 -0
- package/dist/rlm/benchmark.js.map +1 -0
- package/dist/rlm/execute.d.ts +13 -0
- package/dist/rlm/execute.d.ts.map +1 -0
- package/dist/rlm/execute.js +366 -0
- package/dist/rlm/execute.js.map +1 -0
- package/dist/rlm/index.d.ts +6 -0
- package/dist/rlm/index.d.ts.map +1 -0
- package/dist/rlm/index.js +147 -0
- package/dist/rlm/index.js.map +1 -0
- package/dist/rlm/profiles.d.ts +9 -0
- package/dist/rlm/profiles.d.ts.map +1 -0
- package/dist/rlm/profiles.js +46 -0
- package/dist/rlm/profiles.js.map +1 -0
- package/dist/rlm/types.d.ts +98 -0
- package/dist/rlm/types.d.ts.map +1 -0
- package/dist/rlm/types.js +6 -0
- package/dist/rlm/types.js.map +1 -0
- package/dist/rlm/verify.d.ts +13 -0
- package/dist/rlm/verify.d.ts.map +1 -0
- package/dist/rlm/verify.js +58 -0
- package/dist/rlm/verify.js.map +1 -0
- package/dist/rlm/writeback.d.ts +7 -0
- package/dist/rlm/writeback.d.ts.map +1 -0
- package/dist/rlm/writeback.js +77 -0
- package/dist/rlm/writeback.js.map +1 -0
- package/dist/scripts/backfill-memory-embeddings.d.ts +2 -0
- package/dist/scripts/backfill-memory-embeddings.d.ts.map +1 -0
- package/dist/scripts/backfill-memory-embeddings.js +153 -0
- package/dist/scripts/backfill-memory-embeddings.js.map +1 -0
- package/dist/session-buffer.d.ts +104 -0
- package/dist/session-buffer.d.ts.map +1 -0
- package/dist/session-buffer.js +466 -0
- package/dist/session-buffer.js.map +1 -0
- package/dist/session-dedup.d.ts +30 -0
- package/dist/session-dedup.d.ts.map +1 -0
- package/dist/session-dedup.js +67 -0
- package/dist/session-dedup.js.map +1 -0
- package/dist/session-flush.d.ts +81 -0
- package/dist/session-flush.d.ts.map +1 -0
- package/dist/session-flush.js +169 -0
- package/dist/session-flush.js.map +1 -0
- package/dist/session-lifecycle.d.ts +72 -0
- package/dist/session-lifecycle.d.ts.map +1 -0
- package/dist/session-lifecycle.js +247 -0
- package/dist/session-lifecycle.js.map +1 -0
- package/dist/setup.d.ts +2 -0
- package/dist/setup.d.ts.map +1 -0
- package/dist/setup.js +260 -0
- package/dist/setup.js.map +1 -0
- package/dist/stopwords.d.ts +2 -0
- package/dist/stopwords.d.ts.map +1 -0
- package/dist/stopwords.js +20 -0
- package/dist/stopwords.js.map +1 -0
- package/dist/strip-html.d.ts +5 -0
- package/dist/strip-html.d.ts.map +1 -0
- package/dist/strip-html.js +35 -0
- package/dist/strip-html.js.map +1 -0
- package/dist/task-completion-flush.d.ts +36 -0
- package/dist/task-completion-flush.d.ts.map +1 -0
- package/dist/task-completion-flush.js +97 -0
- package/dist/task-completion-flush.js.map +1 -0
- package/dist/task-lifecycle-types.d.ts +13 -0
- package/dist/task-lifecycle-types.d.ts.map +1 -0
- package/dist/task-lifecycle-types.js +12 -0
- package/dist/task-lifecycle-types.js.map +1 -0
- package/dist/task-lifecycle.d.ts +78 -0
- package/dist/task-lifecycle.d.ts.map +1 -0
- package/dist/task-lifecycle.js +256 -0
- package/dist/task-lifecycle.js.map +1 -0
- package/dist/tools/agent-messages.d.ts +26 -0
- package/dist/tools/agent-messages.d.ts.map +1 -0
- package/dist/tools/agent-messages.js +123 -0
- package/dist/tools/agent-messages.js.map +1 -0
- package/dist/tools/agent-tasks.d.ts +24 -0
- package/dist/tools/agent-tasks.d.ts.map +1 -0
- package/dist/tools/agent-tasks.js +162 -0
- package/dist/tools/agent-tasks.js.map +1 -0
- package/dist/tools/archive-memory.d.ts +2 -0
- package/dist/tools/archive-memory.d.ts.map +1 -0
- package/dist/tools/archive-memory.js +19 -0
- package/dist/tools/archive-memory.js.map +1 -0
- package/dist/tools/blind-index.d.ts +29 -0
- package/dist/tools/blind-index.d.ts.map +1 -0
- package/dist/tools/blind-index.js +53 -0
- package/dist/tools/blind-index.js.map +1 -0
- package/dist/tools/cleanup-memories.d.ts +44 -0
- package/dist/tools/cleanup-memories.d.ts.map +1 -0
- package/dist/tools/cleanup-memories.js +126 -0
- package/dist/tools/cleanup-memories.js.map +1 -0
- package/dist/tools/context-checkpoint.d.ts +28 -0
- package/dist/tools/context-checkpoint.d.ts.map +1 -0
- package/dist/tools/context-checkpoint.js +140 -0
- package/dist/tools/context-checkpoint.js.map +1 -0
- package/dist/tools/context-profiles.d.ts +67 -0
- package/dist/tools/context-profiles.d.ts.map +1 -0
- package/dist/tools/context-profiles.js +30 -0
- package/dist/tools/context-profiles.js.map +1 -0
- package/dist/tools/create-note.d.ts +2 -0
- package/dist/tools/create-note.d.ts.map +1 -0
- package/dist/tools/create-note.js +60 -0
- package/dist/tools/create-note.js.map +1 -0
- package/dist/tools/create-vault.d.ts +5 -0
- package/dist/tools/create-vault.d.ts.map +1 -0
- package/dist/tools/create-vault.js +121 -0
- package/dist/tools/create-vault.js.map +1 -0
- package/dist/tools/decrypt-helpers.d.ts +31 -0
- package/dist/tools/decrypt-helpers.d.ts.map +1 -0
- package/dist/tools/decrypt-helpers.js +33 -0
- package/dist/tools/decrypt-helpers.js.map +1 -0
- package/dist/tools/delete-note.d.ts +2 -0
- package/dist/tools/delete-note.d.ts.map +1 -0
- package/dist/tools/delete-note.js +21 -0
- package/dist/tools/delete-note.js.map +1 -0
- package/dist/tools/explore-graph.d.ts +11 -0
- package/dist/tools/explore-graph.d.ts.map +1 -0
- package/dist/tools/explore-graph.js +169 -0
- package/dist/tools/explore-graph.js.map +1 -0
- package/dist/tools/get-related-memories.d.ts +2 -0
- package/dist/tools/get-related-memories.d.ts.map +1 -0
- package/dist/tools/get-related-memories.js +59 -0
- package/dist/tools/get-related-memories.js.map +1 -0
- package/dist/tools/knowledge-links.d.ts +17 -0
- package/dist/tools/knowledge-links.d.ts.map +1 -0
- package/dist/tools/knowledge-links.js +102 -0
- package/dist/tools/knowledge-links.js.map +1 -0
- package/dist/tools/list-active-agents.d.ts +5 -0
- package/dist/tools/list-active-agents.d.ts.map +1 -0
- package/dist/tools/list-active-agents.js +15 -0
- package/dist/tools/list-active-agents.js.map +1 -0
- package/dist/tools/list-notes.d.ts +2 -0
- package/dist/tools/list-notes.d.ts.map +1 -0
- package/dist/tools/list-notes.js +19 -0
- package/dist/tools/list-notes.js.map +1 -0
- package/dist/tools/list-vaults.d.ts +2 -0
- package/dist/tools/list-vaults.d.ts.map +1 -0
- package/dist/tools/list-vaults.js +19 -0
- package/dist/tools/list-vaults.js.map +1 -0
- package/dist/tools/mmr.d.ts +18 -0
- package/dist/tools/mmr.d.ts.map +1 -0
- package/dist/tools/mmr.js +67 -0
- package/dist/tools/mmr.js.map +1 -0
- package/dist/tools/read-memories.d.ts +2 -0
- package/dist/tools/read-memories.d.ts.map +1 -0
- package/dist/tools/read-memories.js +46 -0
- package/dist/tools/read-memories.js.map +1 -0
- package/dist/tools/read-note.d.ts +2 -0
- package/dist/tools/read-note.d.ts.map +1 -0
- package/dist/tools/read-note.js +35 -0
- package/dist/tools/read-note.js.map +1 -0
- package/dist/tools/read-notes.d.ts +6 -0
- package/dist/tools/read-notes.d.ts.map +1 -0
- package/dist/tools/read-notes.js +45 -0
- package/dist/tools/read-notes.js.map +1 -0
- package/dist/tools/resolve-vault-id.d.ts +6 -0
- package/dist/tools/resolve-vault-id.d.ts.map +1 -0
- package/dist/tools/resolve-vault-id.js +7 -0
- package/dist/tools/resolve-vault-id.js.map +1 -0
- package/dist/tools/rrf.d.ts +28 -0
- package/dist/tools/rrf.d.ts.map +1 -0
- package/dist/tools/rrf.js +19 -0
- package/dist/tools/rrf.js.map +1 -0
- package/dist/tools/search-and-read.d.ts +11 -0
- package/dist/tools/search-and-read.d.ts.map +1 -0
- package/dist/tools/search-and-read.js +208 -0
- package/dist/tools/search-and-read.js.map +1 -0
- package/dist/tools/search-memories.d.ts +13 -0
- package/dist/tools/search-memories.d.ts.map +1 -0
- package/dist/tools/search-memories.js +272 -0
- package/dist/tools/search-memories.js.map +1 -0
- package/dist/tools/search-notes.d.ts +2 -0
- package/dist/tools/search-notes.d.ts.map +1 -0
- package/dist/tools/search-notes.js +94 -0
- package/dist/tools/search-notes.js.map +1 -0
- package/dist/tools/semantic-search.d.ts +7 -0
- package/dist/tools/semantic-search.d.ts.map +1 -0
- package/dist/tools/semantic-search.js +85 -0
- package/dist/tools/semantic-search.js.map +1 -0
- package/dist/tools/session-start.d.ts +24 -0
- package/dist/tools/session-start.d.ts.map +1 -0
- package/dist/tools/session-start.js +256 -0
- package/dist/tools/session-start.js.map +1 -0
- package/dist/tools/stale-tasks.d.ts +22 -0
- package/dist/tools/stale-tasks.d.ts.map +1 -0
- package/dist/tools/stale-tasks.js +39 -0
- package/dist/tools/stale-tasks.js.map +1 -0
- package/dist/tools/temporal-decay.d.ts +21 -0
- package/dist/tools/temporal-decay.d.ts.map +1 -0
- package/dist/tools/temporal-decay.js +32 -0
- package/dist/tools/temporal-decay.js.map +1 -0
- package/dist/tools/update-memory.d.ts +19 -0
- package/dist/tools/update-memory.d.ts.map +1 -0
- package/dist/tools/update-memory.js +230 -0
- package/dist/tools/update-memory.js.map +1 -0
- package/dist/tools/update-note.d.ts +2 -0
- package/dist/tools/update-note.d.ts.map +1 -0
- package/dist/tools/update-note.js +79 -0
- package/dist/tools/update-note.js.map +1 -0
- package/dist/tools/vault-instruction-template.d.ts +17 -0
- package/dist/tools/vault-instruction-template.d.ts.map +1 -0
- package/dist/tools/vault-instruction-template.js +77 -0
- package/dist/tools/vault-instruction-template.js.map +1 -0
- package/dist/tools/wiki-link-sync.d.ts +34 -0
- package/dist/tools/wiki-link-sync.d.ts.map +1 -0
- package/dist/tools/wiki-link-sync.js +132 -0
- package/dist/tools/wiki-link-sync.js.map +1 -0
- package/dist/tools/wrap-tool-handler.d.ts +8 -0
- package/dist/tools/wrap-tool-handler.d.ts.map +1 -0
- package/dist/tools/wrap-tool-handler.js +32 -0
- package/dist/tools/wrap-tool-handler.js.map +1 -0
- package/dist/tools/write-memory.d.ts +34 -0
- package/dist/tools/write-memory.d.ts.map +1 -0
- package/dist/tools/write-memory.js +359 -0
- package/dist/tools/write-memory.js.map +1 -0
- package/dist/usage.d.ts +11 -0
- package/dist/usage.d.ts.map +1 -0
- package/dist/usage.js +38 -0
- package/dist/usage.js.map +1 -0
- package/dist/wiki-link-parser.d.ts +27 -0
- package/dist/wiki-link-parser.d.ts.map +1 -0
- package/dist/wiki-link-parser.js +93 -0
- package/dist/wiki-link-parser.js.map +1 -0
- package/package.json +38 -0
|
@@ -0,0 +1,215 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
import { performance } from "node:perf_hooks";
|
|
3
|
+
import { initialize } from "../auth.js";
|
|
4
|
+
import { runController } from "./index.js";
|
|
5
|
+
import { getRlmProfile } from "./profiles.js";
|
|
6
|
+
function parseScenarios() {
|
|
7
|
+
const raw = process.env.RLM_BENCHMARK_SCENARIOS;
|
|
8
|
+
if (!raw) {
|
|
9
|
+
return [
|
|
10
|
+
{
|
|
11
|
+
id: "roadmap",
|
|
12
|
+
goal: "What are the current roadmap priorities and completed milestones for this project?",
|
|
13
|
+
vaultId: process.env.RLM_BENCHMARK_VAULT_ID,
|
|
14
|
+
},
|
|
15
|
+
{
|
|
16
|
+
id: "indexing",
|
|
17
|
+
goal: "What indexing reliability improvements have already been implemented and what remains?",
|
|
18
|
+
vaultId: process.env.RLM_BENCHMARK_VAULT_ID,
|
|
19
|
+
},
|
|
20
|
+
];
|
|
21
|
+
}
|
|
22
|
+
try {
|
|
23
|
+
const parsed = JSON.parse(raw);
|
|
24
|
+
return parsed.filter((s) => s.goal && s.id);
|
|
25
|
+
}
|
|
26
|
+
catch {
|
|
27
|
+
throw new Error("RLM_BENCHMARK_SCENARIOS must be valid JSON array");
|
|
28
|
+
}
|
|
29
|
+
}
|
|
30
|
+
async function main() {
|
|
31
|
+
const ctx = await initialize();
|
|
32
|
+
const envLlmApiKey = process.env.RLM_LLM_API_KEY ??
|
|
33
|
+
process.env.LLM_API_KEY ??
|
|
34
|
+
process.env.DEEPSEEK_API_KEY ??
|
|
35
|
+
undefined;
|
|
36
|
+
const llmConfig = ctx.llmConfig ??
|
|
37
|
+
(envLlmApiKey
|
|
38
|
+
? {
|
|
39
|
+
apiKey: envLlmApiKey,
|
|
40
|
+
baseUrl: process.env.RLM_LLM_BASE_URL ??
|
|
41
|
+
process.env.LLM_BASE_URL ??
|
|
42
|
+
"https://api.deepseek.com",
|
|
43
|
+
modelId: process.env.RLM_LLM_MODEL_ID ??
|
|
44
|
+
process.env.LLM_MODEL_ID ??
|
|
45
|
+
"deepseek-chat",
|
|
46
|
+
}
|
|
47
|
+
: undefined);
|
|
48
|
+
if (!llmConfig) {
|
|
49
|
+
throw new Error("LLM config missing. Add llmApiKey/llmBaseUrl/llmModelId to ~/.exovault-mcp/config.json, or set RLM_LLM_API_KEY in environment.");
|
|
50
|
+
}
|
|
51
|
+
const scenarios = parseScenarios();
|
|
52
|
+
const profileName = process.env.RLM_BENCHMARK_MODE === "quick" ? "quick" : "quality";
|
|
53
|
+
const activeProfile = getRlmProfile(profileName);
|
|
54
|
+
const baseline = {
|
|
55
|
+
profile: profileName,
|
|
56
|
+
maxSteps: Number(process.env.RLM_BENCHMARK_BASELINE_MAX_STEPS ?? activeProfile.maxSteps),
|
|
57
|
+
memorySearchThreshold: Number(process.env.RLM_BENCHMARK_BASELINE_THRESHOLD ?? activeProfile.tuning.memorySearchThreshold),
|
|
58
|
+
memorySearchTopK: Number(process.env.RLM_BENCHMARK_BASELINE_MEMORY_TOP_K ?? activeProfile.tuning.memorySearchTopK),
|
|
59
|
+
noteSearchTopK: Number(process.env.RLM_BENCHMARK_BASELINE_NOTE_TOP_K ?? activeProfile.tuning.noteSearchTopK),
|
|
60
|
+
stepResultCharLimit: Number(process.env.RLM_BENCHMARK_BASELINE_STEP_RESULT_LIMIT ??
|
|
61
|
+
activeProfile.tuning.stepResultCharLimit),
|
|
62
|
+
summaryPerStepCharLimit: Number(process.env.RLM_BENCHMARK_BASELINE_SUMMARY_STEP_LIMIT ??
|
|
63
|
+
activeProfile.tuning.summaryPerStepCharLimit),
|
|
64
|
+
duplicateSearchTolerance: Number(process.env.RLM_BENCHMARK_BASELINE_DUPLICATE_TOLERANCE ??
|
|
65
|
+
activeProfile.tuning.duplicateSearchTolerance),
|
|
66
|
+
forceFinalizeRemainingSteps: Number(process.env.RLM_BENCHMARK_BASELINE_FORCE_FINALIZE_REMAINING ??
|
|
67
|
+
activeProfile.tuning.forceFinalizeRemainingSteps),
|
|
68
|
+
forceFinalizeMinEvidence: Number(process.env.RLM_BENCHMARK_BASELINE_FORCE_FINALIZE_EVIDENCE ??
|
|
69
|
+
activeProfile.tuning.forceFinalizeMinEvidence),
|
|
70
|
+
fallbackRecencyRetryLimit: Number(process.env.RLM_BENCHMARK_BASELINE_FALLBACK_RECENCY_RETRY_LIMIT ??
|
|
71
|
+
activeProfile.tuning.fallbackRecencyRetryLimit),
|
|
72
|
+
disableLlmFallbackSummary: (process.env.RLM_BENCHMARK_BASELINE_DISABLE_LLM_FALLBACK_SUMMARY ??
|
|
73
|
+
String(activeProfile.tuning.disableLlmFallbackSummary)) === "true",
|
|
74
|
+
};
|
|
75
|
+
const tuned = {
|
|
76
|
+
profile: profileName,
|
|
77
|
+
maxSteps: Number(process.env.RLM_BENCHMARK_TUNED_MAX_STEPS ?? Math.max(2, activeProfile.maxSteps - 1)),
|
|
78
|
+
memorySearchThreshold: Number(process.env.RLM_BENCHMARK_TUNED_THRESHOLD ?? activeProfile.tuning.memorySearchThreshold),
|
|
79
|
+
memorySearchTopK: Number(process.env.RLM_BENCHMARK_TUNED_MEMORY_TOP_K ?? Math.max(2, activeProfile.tuning.memorySearchTopK - 1)),
|
|
80
|
+
noteSearchTopK: Number(process.env.RLM_BENCHMARK_TUNED_NOTE_TOP_K ?? activeProfile.tuning.noteSearchTopK),
|
|
81
|
+
stepResultCharLimit: Number(process.env.RLM_BENCHMARK_TUNED_STEP_RESULT_LIMIT ?? activeProfile.tuning.stepResultCharLimit),
|
|
82
|
+
summaryPerStepCharLimit: Number(process.env.RLM_BENCHMARK_TUNED_SUMMARY_STEP_LIMIT ??
|
|
83
|
+
activeProfile.tuning.summaryPerStepCharLimit),
|
|
84
|
+
duplicateSearchTolerance: Number(process.env.RLM_BENCHMARK_TUNED_DUPLICATE_TOLERANCE ??
|
|
85
|
+
activeProfile.tuning.duplicateSearchTolerance),
|
|
86
|
+
forceFinalizeRemainingSteps: Number(process.env.RLM_BENCHMARK_TUNED_FORCE_FINALIZE_REMAINING ??
|
|
87
|
+
activeProfile.tuning.forceFinalizeRemainingSteps),
|
|
88
|
+
forceFinalizeMinEvidence: Number(process.env.RLM_BENCHMARK_TUNED_FORCE_FINALIZE_EVIDENCE ??
|
|
89
|
+
activeProfile.tuning.forceFinalizeMinEvidence),
|
|
90
|
+
fallbackRecencyRetryLimit: Number(process.env.RLM_BENCHMARK_TUNED_FALLBACK_RECENCY_RETRY_LIMIT ??
|
|
91
|
+
activeProfile.tuning.fallbackRecencyRetryLimit),
|
|
92
|
+
disableLlmFallbackSummary: (process.env.RLM_BENCHMARK_TUNED_DISABLE_LLM_FALLBACK_SUMMARY ??
|
|
93
|
+
String(activeProfile.tuning.disableLlmFallbackSummary)) === "true",
|
|
94
|
+
};
|
|
95
|
+
const guardrailPolicy = {
|
|
96
|
+
tunedMaxSteps: Number(process.env.RLM_BENCHMARK_ASSERT_TUNED_MAX_STEPS ?? tuned.maxSteps),
|
|
97
|
+
tunedMaxTokens: process.env.RLM_BENCHMARK_ASSERT_TUNED_MAX_TOKENS
|
|
98
|
+
? Number(process.env.RLM_BENCHMARK_ASSERT_TUNED_MAX_TOKENS)
|
|
99
|
+
: undefined,
|
|
100
|
+
tunedMaxFallbackOnlyRate: Number(process.env.RLM_BENCHMARK_ASSERT_TUNED_MAX_FALLBACK_ONLY_RATE ?? 0.2),
|
|
101
|
+
strict: process.env.RLM_BENCHMARK_STRICT_ASSERT === "true",
|
|
102
|
+
};
|
|
103
|
+
const guardrailChecks = [];
|
|
104
|
+
const runInParallel = process.env.RLM_BENCHMARK_PARALLEL !== "false";
|
|
105
|
+
const report = {
|
|
106
|
+
generatedAt: new Date().toISOString(),
|
|
107
|
+
model: llmConfig.modelId,
|
|
108
|
+
benchmarkMode: profileName,
|
|
109
|
+
profiles: { baseline, tuned },
|
|
110
|
+
guardrails: guardrailPolicy,
|
|
111
|
+
parallel: runInParallel,
|
|
112
|
+
scenarios: [],
|
|
113
|
+
};
|
|
114
|
+
for (const scenario of scenarios) {
|
|
115
|
+
const scenarioResult = {
|
|
116
|
+
id: scenario.id,
|
|
117
|
+
goal: scenario.goal,
|
|
118
|
+
runs: [],
|
|
119
|
+
};
|
|
120
|
+
const modes = ["baseline", "tuned"];
|
|
121
|
+
const executeMode = async (mode) => {
|
|
122
|
+
const cfg = mode === "baseline" ? baseline : tuned;
|
|
123
|
+
const start = performance.now();
|
|
124
|
+
const run = await runController(ctx, llmConfig, {
|
|
125
|
+
goal: scenario.goal,
|
|
126
|
+
vaultId: scenario.vaultId,
|
|
127
|
+
profile: cfg.profile,
|
|
128
|
+
maxSteps: cfg.maxSteps,
|
|
129
|
+
writebackEnabled: false,
|
|
130
|
+
memorySearchThreshold: cfg.memorySearchThreshold,
|
|
131
|
+
memorySearchTopK: cfg.memorySearchTopK,
|
|
132
|
+
noteSearchTopK: cfg.noteSearchTopK,
|
|
133
|
+
disableLlmFallbackSummary: cfg.disableLlmFallbackSummary,
|
|
134
|
+
stepResultCharLimit: cfg.stepResultCharLimit,
|
|
135
|
+
summaryPerStepCharLimit: cfg.summaryPerStepCharLimit,
|
|
136
|
+
duplicateSearchTolerance: cfg.duplicateSearchTolerance,
|
|
137
|
+
forceFinalizeRemainingSteps: cfg.forceFinalizeRemainingSteps,
|
|
138
|
+
forceFinalizeMinEvidence: cfg.forceFinalizeMinEvidence,
|
|
139
|
+
fallbackRecencyRetryLimit: cfg.fallbackRecencyRetryLimit,
|
|
140
|
+
});
|
|
141
|
+
const elapsed = Math.round(performance.now() - start);
|
|
142
|
+
const memorySearchCalls = run.metrics?.memorySearchCalls ?? 0;
|
|
143
|
+
const memoryFallbackCalls = run.metrics?.memoryFallbackCalls ?? 0;
|
|
144
|
+
const fallbackOnly = memorySearchCalls > 0 && memoryFallbackCalls >= memorySearchCalls;
|
|
145
|
+
const result = {
|
|
146
|
+
mode,
|
|
147
|
+
elapsedMs: elapsed,
|
|
148
|
+
answer: run.answer,
|
|
149
|
+
answerPreview: run.answer.slice(0, 500),
|
|
150
|
+
confidence: run.confidence,
|
|
151
|
+
stepsUsed: run.stepsUsed,
|
|
152
|
+
totalTokens: run.totalTokens,
|
|
153
|
+
finishReason: run.finishReason,
|
|
154
|
+
budgetExhausted: run.budgetExhausted,
|
|
155
|
+
fallbackOnly,
|
|
156
|
+
metrics: run.metrics,
|
|
157
|
+
};
|
|
158
|
+
if (mode === "tuned") {
|
|
159
|
+
guardrailChecks.push({
|
|
160
|
+
scenarioId: scenario.id,
|
|
161
|
+
mode,
|
|
162
|
+
passed: result.stepsUsed <= guardrailPolicy.tunedMaxSteps,
|
|
163
|
+
message: `stepsUsed=${result.stepsUsed}, ceiling=${guardrailPolicy.tunedMaxSteps}`,
|
|
164
|
+
});
|
|
165
|
+
if (guardrailPolicy.tunedMaxTokens !== undefined) {
|
|
166
|
+
guardrailChecks.push({
|
|
167
|
+
scenarioId: scenario.id,
|
|
168
|
+
mode,
|
|
169
|
+
passed: result.totalTokens <= guardrailPolicy.tunedMaxTokens,
|
|
170
|
+
message: `totalTokens=${result.totalTokens}, ceiling=${guardrailPolicy.tunedMaxTokens}`,
|
|
171
|
+
});
|
|
172
|
+
}
|
|
173
|
+
const acceptableFinish = result.finishReason === "tool_finish" || result.finishReason === "controller_finalize";
|
|
174
|
+
guardrailChecks.push({
|
|
175
|
+
scenarioId: scenario.id,
|
|
176
|
+
mode,
|
|
177
|
+
passed: acceptableFinish,
|
|
178
|
+
message: `finishReason=${result.finishReason ?? "none"} expected tool_finish/controller_finalize`,
|
|
179
|
+
});
|
|
180
|
+
const fallbackOnlyRate = fallbackOnly ? 1 : 0;
|
|
181
|
+
guardrailChecks.push({
|
|
182
|
+
scenarioId: scenario.id,
|
|
183
|
+
mode,
|
|
184
|
+
passed: fallbackOnlyRate <= guardrailPolicy.tunedMaxFallbackOnlyRate,
|
|
185
|
+
message: `fallbackOnlyRate=${fallbackOnlyRate}, ceiling=${guardrailPolicy.tunedMaxFallbackOnlyRate}`,
|
|
186
|
+
});
|
|
187
|
+
}
|
|
188
|
+
return result;
|
|
189
|
+
};
|
|
190
|
+
const runResults = runInParallel
|
|
191
|
+
? await Promise.all(modes.map((mode) => executeMode(mode)))
|
|
192
|
+
: [
|
|
193
|
+
await executeMode("baseline"),
|
|
194
|
+
await executeMode("tuned"),
|
|
195
|
+
];
|
|
196
|
+
scenarioResult.runs.push(...runResults);
|
|
197
|
+
report.scenarios.push(scenarioResult);
|
|
198
|
+
}
|
|
199
|
+
report.checks = guardrailChecks;
|
|
200
|
+
const failedChecks = guardrailChecks.filter((check) => !check.passed);
|
|
201
|
+
report.checkSummary = {
|
|
202
|
+
total: guardrailChecks.length,
|
|
203
|
+
failed: failedChecks.length,
|
|
204
|
+
};
|
|
205
|
+
if (guardrailPolicy.strict && failedChecks.length > 0) {
|
|
206
|
+
throw new Error(`Benchmark guardrails failed: ${failedChecks
|
|
207
|
+
.map((check) => `${check.scenarioId}/${check.mode} ${check.message}`)
|
|
208
|
+
.join("; ")}`);
|
|
209
|
+
}
|
|
210
|
+
process.stdout.write(`${JSON.stringify(report, null, 2)}\n`);
|
|
211
|
+
}
|
|
212
|
+
main().catch((err) => {
|
|
213
|
+
process.stderr.write(`RLM benchmark failed: ${err.message}\n`);
|
|
214
|
+
process.exit(1);
|
|
215
|
+
});
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark.js","sourceRoot":"","sources":["../../src/rlm/benchmark.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAE3C,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAiB9C,SAAS,cAAc;IACrB,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC;IAChD,IAAI,CAAC,GAAG,EAAE,CAAC;QACT,OAAO;YACL;gBACE,EAAE,EAAE,SAAS;gBACb,IAAI,EAAE,oFAAoF;gBAC1F,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,sBAAsB;aAC5C;YACD;gBACE,EAAE,EAAE,UAAU;gBACd,IAAI,EAAE,wFAAwF;gBAC9F,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,sBAAsB;aAC5C;SACF,CAAC;IACJ,CAAC;IACD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAe,CAAC;QAC7C,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC;IAC9C,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;IACtE,CAAC;AACH,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,GAAG,GAAG,MAAM,UAAU,EAAE,CAAC;IAC/B,MAAM,YAAY,GAChB,OAAO,CAAC,GAAG,CAAC,eAAe;QAC3B,OAAO,CAAC,GAAG,CAAC,WAAW;QACvB,OAAO,CAAC,GAAG,CAAC,gBAAgB;QAC5B,SAAS,CAAC;IACZ,MAAM,SAAS,GACb,GAAG,CAAC,SAAS;QACb,CAAC,YAAY;YACX,CAAC,CAAC;gBACE,MAAM,EAAE,YAAY;gBACpB,OAAO,EACL,OAAO,CAAC,GAAG,CAAC,gBAAgB;oBAC5B,OAAO,CAAC,GAAG,CAAC,YAAY;oBACxB,0BAA0B;gBAC5B,OAAO,EACL,OAAO,CAAC,GAAG,CAAC,gBAAgB;oBAC5B,OAAO,CAAC,GAAG,CAAC,YAAY;oBACxB,eAAe;aAClB;YACH,CAAC,CAAC,SAAS,CAAC,CAAC;IACjB,IAAI,CAAC,SAAS,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CACb,gIAAgI,CACjI,CAAC;IACJ,CAAC;IAED,MAAM,SAAS,GAAG,cAAc,EAAE,CAAC;IACnC,MAAM,WAAW,GACf,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;IACnE,MAAM,aAAa,GAAG,aAAa,CAAC,WAAW,CAAC,CAAC;IACjD,MAAM,QAAQ,GAAG;QACf,OAAO,EAAE,WAAW;QACpB,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,gCAAgC,IAAI,aAAa,CAAC,QAAQ,CAAC;QACxF,qBAAqB,EAAE,MAAM,CAC3B,OAAO,CAAC,GAAG,CAAC,gCAAgC,IAAI,aAAa,CAAC,MAAM,CAAC,qBAAqB,CAC3F;QACD,gBAAgB,EAAE,MAAM,CACtB,OAAO,CAAC,GAAG,CAAC,mCAAmC,IAAI,aAAa,CAAC,MAAM,CAAC,gBAAgB,CACzF;QACD,cAAc,EAAE,MAAM,CACpB,OAAO,CAAC,GAAG,CAAC,iCAAiC,IAAI,aAAa,CAAC,MAAM,CAAC,cAAc,CACrF;QACD,mBAAmB,EAAE,MAAM,CACzB,OAAO,CAAC,GAAG,CAAC,wCAAwC;YAClD,aAAa,CAAC,MAAM,CAAC,mBAAmB,CAC3C;QACD,uBAAuB,EAAE,MAAM,CAC7B,OAAO,CAAC,GAAG,CAAC,yCAAyC;YACnD,aAAa,CAAC,MAAM,CAAC,uBAAuB,CAC/C;QACD,wBAAwB,EAAE,MAAM,CAC9B,OAAO,CAAC,GAAG,CAAC,0CAA0C;YACpD,aAAa,CAAC,MAAM,CAAC,wBAAwB,CAChD;QACD,2BAA2B,EAAE,MAAM,CACjC,OAAO,CAAC,GAAG,CAAC,+CAA+C;YACzD,aAAa,CAAC,MAAM,CAAC,2BAA2B,CACnD;QACD,wBAAwB,EAAE,MAAM,CAC9B,OAAO,CAAC,GAAG,CAAC,8CAA8C;YACxD,aAAa,CAAC,MAAM,CAAC,wBAAwB,CAChD;QACD,yBAAyB,EAAE,MAAM,CAC/B,OAAO,CAAC,GAAG,CAAC,mDAAmD;YAC7D,aAAa,CAAC,MAAM,CAAC,yBAAyB,CACjD;QACD,yBAAyB,EACvB,CAAC,OAAO,CAAC,GAAG,CAAC,mDAAmD;YAC9D,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,yBAAyB,CAAC,CAAC,KAAK,MAAM;KACvE,CAAC;IACF,MAAM,KAAK,GAAG;QACZ,OAAO,EAAE,WAAW;QACpB,QAAQ,EAAE,MAAM,CACd,OAAO,CAAC,GAAG,CAAC,6BAA6B,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,aAAa,CAAC,QAAQ,GAAG,CAAC,CAAC,CACrF;QACD,qBAAqB,EAAE,MAAM,CAC3B,OAAO,CAAC,GAAG,CAAC,6BAA6B,IAAI,aAAa,CAAC,MAAM,CAAC,qBAAqB,CACxF;QACD,gBAAgB,EAAE,MAAM,CACtB,OAAO,CAAC,GAAG,CAAC,gCAAgC,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,aAAa,CAAC,MAAM,CAAC,gBAAgB,GAAG,CAAC,CAAC,CACvG;QACD,cAAc,EAAE,MAAM,CACpB,OAAO,CAAC,GAAG,CAAC,8BAA8B,IAAI,aAAa,CAAC,MAAM,CAAC,cAAc,CAClF;QACD,mBAAmB,EAAE,MAAM,CACzB,OAAO,CAAC,GAAG,CAAC,qCAAqC,IAAI,aAAa,CAAC,MAAM,CAAC,mBAAmB,CAC9F;QACD,uBAAuB,EAAE,MAAM,CAC7B,OAAO,CAAC,GAAG,CAAC,sCAAsC;YAChD,aAAa,CAAC,MAAM,CAAC,uBAAuB,CAC/C;QACD,wBAAwB,EAAE,MAAM,CAC9B,OAAO,CAAC,GAAG,CAAC,uCAAuC;YACjD,aAAa,CAAC,MAAM,CAAC,wBAAwB,CAChD;QACD,2BAA2B,EAAE,MAAM,CACjC,OAAO,CAAC,GAAG,CAAC,4CAA4C;YACtD,aAAa,CAAC,MAAM,CAAC,2BAA2B,CACnD;QACD,wBAAwB,EAAE,MAAM,CAC9B,OAAO,CAAC,GAAG,CAAC,2CAA2C;YACrD,aAAa,CAAC,MAAM,CAAC,wBAAwB,CAChD;QACD,yBAAyB,EAAE,MAAM,CAC/B,OAAO,CAAC,GAAG,CAAC,gDAAgD;YAC1D,aAAa,CAAC,MAAM,CAAC,yBAAyB,CACjD;QACD,yBAAyB,EACvB,CAAC,OAAO,CAAC,GAAG,CAAC,gDAAgD;YAC3D,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,yBAAyB,CAAC,CAAC,KAAK,MAAM;KACvE,CAAC;IACF,MAAM,eAAe,GAAG;QACtB,aAAa,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,oCAAoC,IAAI,KAAK,CAAC,QAAQ,CAAC;QACzF,cAAc,EAAE,OAAO,CAAC,GAAG,CAAC,qCAAqC;YAC/D,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,qCAAqC,CAAC;YAC3D,CAAC,CAAC,SAAS;QACb,wBAAwB,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,iDAAiD,IAAI,GAAG,CAAC;QACtG,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,2BAA2B,KAAK,MAAM;KAC3D,CAAC;IACF,MAAM,eAAe,GAAqB,EAAE,CAAC;IAC7C,MAAM,aAAa,GAAG,OAAO,CAAC,GAAG,CAAC,sBAAsB,KAAK,OAAO,CAAC;IAErE,MAAM,MAAM,GAA4B;QACtC,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACrC,KAAK,EAAE,SAAS,CAAC,OAAO;QACxB,aAAa,EAAE,WAAW;QAC1B,QAAQ,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE;QAC7B,UAAU,EAAE,eAAe;QAC3B,QAAQ,EAAE,aAAa;QACvB,SAAS,EAAE,EAAE;KACd,CAAC;IAEF,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;QACjC,MAAM,cAAc,GAA4B;YAC9C,EAAE,EAAE,QAAQ,CAAC,EAAE;YACf,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,IAAI,EAAE,EAAE;SACT,CAAC;QAEF,MAAM,KAAK,GAAG,CAAC,UAAU,EAAE,OAAO,CAAU,CAAC;QAC7C,MAAM,WAAW,GAAG,KAAK,EAAE,IAAmB,EAAE,EAAE;YAC9C,MAAM,GAAG,GAAG,IAAI,KAAK,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC;YACnD,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC;YAChC,MAAM,GAAG,GAAG,MAAM,aAAa,CAAC,GAAG,EAAE,SAAS,EAAE;gBAC9C,IAAI,EAAE,QAAQ,CAAC,IAAI;gBACnB,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,OAAO,EAAE,GAAG,CAAC,OAAO;gBACpB,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACtB,gBAAgB,EAAE,KAAK;gBACvB,qBAAqB,EAAE,GAAG,CAAC,qBAAqB;gBAChD,gBAAgB,EAAE,GAAG,CAAC,gBAAgB;gBACtC,cAAc,EAAE,GAAG,CAAC,cAAc;gBAClC,yBAAyB,EAAE,GAAG,CAAC,yBAAyB;gBACxD,mBAAmB,EAAE,GAAG,CAAC,mBAAmB;gBAC5C,uBAAuB,EAAE,GAAG,CAAC,uBAAuB;gBACpD,wBAAwB,EAAE,GAAG,CAAC,wBAAwB;gBACtD,2BAA2B,EAAE,GAAG,CAAC,2BAA2B;gBAC5D,wBAAwB,EAAE,GAAG,CAAC,wBAAwB;gBACtD,yBAAyB,EAAE,GAAG,CAAC,yBAAyB;aACzD,CAAC,CAAC;YACH,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,CAAC;YACtD,MAAM,iBAAiB,GAAG,GAAG,CAAC,OAAO,EAAE,iBAAiB,IAAI,CAAC,CAAC;YAC9D,MAAM,mBAAmB,GAAG,GAAG,CAAC,OAAO,EAAE,mBAAmB,IAAI,CAAC,CAAC;YAClE,MAAM,YAAY,GAChB,iBAAiB,GAAG,CAAC,IAAI,mBAAmB,IAAI,iBAAiB,CAAC;YACpE,MAAM,MAAM,GAAG;gBACb,IAAI;gBACJ,SAAS,EAAE,OAAO;gBAClB,MAAM,EAAE,GAAG,CAAC,MAAM;gBAClB,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBACvC,UAAU,EAAE,GAAG,CAAC,UAAU;gBAC1B,SAAS,EAAE,GAAG,CAAC,SAAS;gBACxB,WAAW,EAAE,GAAG,CAAC,WAAW;gBAC5B,YAAY,EAAE,GAAG,CAAC,YAAY;gBAC9B,eAAe,EAAE,GAAG,CAAC,eAAe;gBACpC,YAAY;gBACZ,OAAO,EAAE,GAAG,CAAC,OAAO;aACrB,CAAC;YACF,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;gBACrB,eAAe,CAAC,IAAI,CAAC;oBACnB,UAAU,EAAE,QAAQ,CAAC,EAAE;oBACvB,IAAI;oBACJ,MAAM,EAAE,MAAM,CAAC,SAAS,IAAI,eAAe,CAAC,aAAa;oBACzD,OAAO,EAAE,aAAa,MAAM,CAAC,SAAS,aAAa,eAAe,CAAC,aAAa,EAAE;iBACnF,CAAC,CAAC;gBACH,IAAI,eAAe,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;oBACjD,eAAe,CAAC,IAAI,CAAC;wBACnB,UAAU,EAAE,QAAQ,CAAC,EAAE;wBACvB,IAAI;wBACJ,MAAM,EAAE,MAAM,CAAC,WAAW,IAAI,eAAe,CAAC,cAAc;wBAC5D,OAAO,EAAE,eAAe,MAAM,CAAC,WAAW,aAAa,eAAe,CAAC,cAAc,EAAE;qBACxF,CAAC,CAAC;gBACL,CAAC;gBACD,MAAM,gBAAgB,GACpB,MAAM,CAAC,YAAY,KAAK,aAAa,IAAI,MAAM,CAAC,YAAY,KAAK,qBAAqB,CAAC;gBACzF,eAAe,CAAC,IAAI,CAAC;oBACnB,UAAU,EAAE,QAAQ,CAAC,EAAE;oBACvB,IAAI;oBACJ,MAAM,EAAE,gBAAgB;oBACxB,OAAO,EAAE,gBAAgB,MAAM,CAAC,YAAY,IAAI,MAAM,2CAA2C;iBAClG,CAAC,CAAC;gBACH,MAAM,gBAAgB,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBAC9C,eAAe,CAAC,IAAI,CAAC;oBACnB,UAAU,EAAE,QAAQ,CAAC,EAAE;oBACvB,IAAI;oBACJ,MAAM,EAAE,gBAAgB,IAAI,eAAe,CAAC,wBAAwB;oBACpE,OAAO,EACL,oBAAoB,gBAAgB,aAAa,eAAe,CAAC,wBAAwB,EAAE;iBAC9F,CAAC,CAAC;YACL,CAAC;YACD,OAAO,MAAM,CAAC;QAClB,CAAC,CAAC;QACF,MAAM,UAAU,GAAG,aAAa;YAC9B,CAAC,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC;YAC3D,CAAC,CAAC;gBACE,MAAM,WAAW,CAAC,UAAU,CAAC;gBAC7B,MAAM,WAAW,CAAC,OAAO,CAAC;aAC3B,CAAC;QACL,cAAc,CAAC,IAAkB,CAAC,IAAI,CAAC,GAAG,UAAU,CAAC,CAAC;QAEtD,MAAM,CAAC,SAAuB,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;IACvD,CAAC;IACD,MAAM,CAAC,MAAM,GAAG,eAAe,CAAC;IAChC,MAAM,YAAY,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACtE,MAAM,CAAC,YAAY,GAAG;QACpB,KAAK,EAAE,eAAe,CAAC,MAAM;QAC7B,MAAM,EAAE,YAAY,CAAC,MAAM;KAC5B,CAAC;IACF,IAAI,eAAe,CAAC,MAAM,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtD,MAAM,IAAI,KAAK,CACb,gCAAgC,YAAY;aACzC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,GAAG,KAAK,CAAC,UAAU,IAAI,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;aACpE,IAAI,CAAC,IAAI,CAAC,EAAE,CAChB,CAAC;IACJ,CAAC;IAED,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;AAC/D,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACnB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,yBAA0B,GAAa,CAAC,OAAO,IAAI,CAAC,CAAC;IAC1E,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import type { McpContext, LlmConfig } from "../auth.js";
|
|
2
|
+
import type { ControllerState } from "./types.js";
|
|
3
|
+
export interface ExecuteResult {
|
|
4
|
+
answer: string;
|
|
5
|
+
confidence: number;
|
|
6
|
+
state: ControllerState;
|
|
7
|
+
}
|
|
8
|
+
/**
|
|
9
|
+
* Run the recursive execution loop.
|
|
10
|
+
* Uses Vercel AI SDK generateText() with tool definitions.
|
|
11
|
+
* The LLM proposes actions, we execute them, and loop until finish() or budget exhaustion.
|
|
12
|
+
*/
|
|
13
|
+
export declare function executeLoop(ctx: McpContext, llmConfig: LlmConfig, state: ControllerState): Promise<ExecuteResult>;
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"execute.d.ts","sourceRoot":"","sources":["../../src/rlm/execute.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAExD,OAAO,KAAK,EAAE,eAAe,EAAW,MAAM,YAAY,CAAC;AA2D3D,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,eAAe,CAAC;CACxB;AAED;;;;GAIG;AACH,wBAAsB,WAAW,CAC/B,GAAG,EAAE,UAAU,EACf,SAAS,EAAE,SAAS,EACpB,KAAK,EAAE,eAAe,GACrB,OAAO,CAAC,aAAa,CAAC,CA0VxB"}
|
|
@@ -0,0 +1,366 @@
|
|
|
1
|
+
import { generateText, stepCountIs } from "ai";
|
|
2
|
+
import { createOpenAI } from "@ai-sdk/openai";
|
|
3
|
+
import { buildToolDefinitions } from "./actions.js";
|
|
4
|
+
const SYSTEM_PROMPT = `You are a retrieval controller for ExoVault, an encrypted knowledge base.
|
|
5
|
+
Your job is to answer a goal by searching and reading memories, then calling finish().
|
|
6
|
+
|
|
7
|
+
## Optimal workflow (aim for 3-4 steps total):
|
|
8
|
+
1. search_memories — one good semantic query covers most goals
|
|
9
|
+
2. read_memories — read 3-5 top hits for full details (ONLY if summaries are insufficient)
|
|
10
|
+
3. finish() — synthesize your answer with confidence score
|
|
11
|
+
|
|
12
|
+
## CRITICAL RULES
|
|
13
|
+
- You have a STRICT step budget. Every step costs money.
|
|
14
|
+
- call finish() AS SOON AS you have enough evidence. Do not over-search.
|
|
15
|
+
- ONE search_memories call usually suffices. Only search again if the first returned zero relevant hits.
|
|
16
|
+
- Do NOT call search_notes unless you specifically need keyword matching on note titles/tags. Most goals are answered by memories alone.
|
|
17
|
+
- Do NOT repeat searches with slightly different wording — read the results you already have.
|
|
18
|
+
- Do NOT call finish() more than once.
|
|
19
|
+
- If no relevant results after 1-2 searches, call finish() with low confidence immediately.
|
|
20
|
+
- The finish() tool is the ONLY way to return your answer. Text responses are NOT returned.`;
|
|
21
|
+
function buildUserMessage(state) {
|
|
22
|
+
const parts = [`Goal: ${state.goal}`];
|
|
23
|
+
if (state.hypothesis) {
|
|
24
|
+
parts.push(`\nCurrent hypothesis: ${state.hypothesis}`);
|
|
25
|
+
}
|
|
26
|
+
if (state.retrieved.length > 0) {
|
|
27
|
+
parts.push(`\nRetrieved so far (${state.retrieved.length} items):`);
|
|
28
|
+
for (const item of state.retrieved.slice(-10)) {
|
|
29
|
+
parts.push(`- [${item.type}:${item.id}] ${item.summary}`);
|
|
30
|
+
}
|
|
31
|
+
}
|
|
32
|
+
if (state.steps.length > 0) {
|
|
33
|
+
parts.push(`\nSteps taken: ${state.steps.length}/${state.budget.maxSteps}`);
|
|
34
|
+
}
|
|
35
|
+
const stepsRemaining = state.budget.maxSteps - state.steps.length;
|
|
36
|
+
if (stepsRemaining <= state.tuning.forceFinalizeRemainingSteps + 1) {
|
|
37
|
+
parts.push(`\nWARNING: Step budget is low (${stepsRemaining} steps left). If evidence is sufficient, call finish() now.`);
|
|
38
|
+
}
|
|
39
|
+
const elapsed = Date.now() - state.startedAt;
|
|
40
|
+
const remaining = state.budget.maxWallMs - elapsed;
|
|
41
|
+
if (remaining < state.budget.maxWallMs * 0.3) {
|
|
42
|
+
parts.push(`\nWARNING: Time running low (${Math.round(remaining / 1000)}s remaining). Finish soon.`);
|
|
43
|
+
}
|
|
44
|
+
return parts.join("\n");
|
|
45
|
+
}
|
|
46
|
+
function isBudgetExhausted(state) {
|
|
47
|
+
if (state.steps.length >= state.budget.maxSteps)
|
|
48
|
+
return true;
|
|
49
|
+
if (Date.now() - state.startedAt >= state.budget.maxWallMs)
|
|
50
|
+
return true;
|
|
51
|
+
return false;
|
|
52
|
+
}
|
|
53
|
+
/**
|
|
54
|
+
* Run the recursive execution loop.
|
|
55
|
+
* Uses Vercel AI SDK generateText() with tool definitions.
|
|
56
|
+
* The LLM proposes actions, we execute them, and loop until finish() or budget exhaustion.
|
|
57
|
+
*/
|
|
58
|
+
export async function executeLoop(ctx, llmConfig, state) {
|
|
59
|
+
const provider = createOpenAI({
|
|
60
|
+
apiKey: llmConfig.apiKey,
|
|
61
|
+
baseURL: llmConfig.baseUrl,
|
|
62
|
+
});
|
|
63
|
+
const model = provider.chat(llmConfig.modelId);
|
|
64
|
+
const tools = buildToolDefinitions(ctx, state, state.vaultId);
|
|
65
|
+
// Calculate remaining steps for this loop
|
|
66
|
+
const remainingSteps = state.budget.maxSteps - state.steps.length;
|
|
67
|
+
const disableLlmFallbackSummary = state.tuning.disableLlmFallbackSummary;
|
|
68
|
+
function clip(text, limit) {
|
|
69
|
+
if (text.length <= limit)
|
|
70
|
+
return text;
|
|
71
|
+
return `${text.slice(0, limit)}…`;
|
|
72
|
+
}
|
|
73
|
+
function clampConfidence(value) {
|
|
74
|
+
if (typeof value !== "number" || Number.isNaN(value))
|
|
75
|
+
return undefined;
|
|
76
|
+
return Math.max(0, Math.min(1, value));
|
|
77
|
+
}
|
|
78
|
+
function parseFinalizePayload(text) {
|
|
79
|
+
try {
|
|
80
|
+
const parsed = JSON.parse(text);
|
|
81
|
+
return {
|
|
82
|
+
answer: typeof parsed.answer === "string" ? parsed.answer : undefined,
|
|
83
|
+
confidence: clampConfidence(parsed.confidence),
|
|
84
|
+
};
|
|
85
|
+
}
|
|
86
|
+
catch {
|
|
87
|
+
const firstBrace = text.indexOf("{");
|
|
88
|
+
const lastBrace = text.lastIndexOf("}");
|
|
89
|
+
if (firstBrace >= 0 && lastBrace > firstBrace) {
|
|
90
|
+
try {
|
|
91
|
+
const parsed = JSON.parse(text.slice(firstBrace, lastBrace + 1));
|
|
92
|
+
return {
|
|
93
|
+
answer: typeof parsed.answer === "string" ? parsed.answer : undefined,
|
|
94
|
+
confidence: clampConfidence(parsed.confidence),
|
|
95
|
+
};
|
|
96
|
+
}
|
|
97
|
+
catch {
|
|
98
|
+
return null;
|
|
99
|
+
}
|
|
100
|
+
}
|
|
101
|
+
return null;
|
|
102
|
+
}
|
|
103
|
+
}
|
|
104
|
+
function parseToolOutput(raw) {
|
|
105
|
+
if (typeof raw !== "string")
|
|
106
|
+
return raw;
|
|
107
|
+
try {
|
|
108
|
+
return JSON.parse(raw);
|
|
109
|
+
}
|
|
110
|
+
catch {
|
|
111
|
+
return raw;
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
function buildResultSnapshot(action, parsed) {
|
|
115
|
+
if (!parsed || typeof parsed !== "object")
|
|
116
|
+
return undefined;
|
|
117
|
+
const obj = parsed;
|
|
118
|
+
if (action === "search_memories") {
|
|
119
|
+
const memories = Array.isArray(obj.memories) ? obj.memories : [];
|
|
120
|
+
const hits = memories.slice(0, 5).map((m) => {
|
|
121
|
+
const mm = m;
|
|
122
|
+
return {
|
|
123
|
+
id: mm.id,
|
|
124
|
+
memoryType: mm.memoryType,
|
|
125
|
+
summary: typeof mm.summary === "string" ? clip(mm.summary, 120) : null,
|
|
126
|
+
similarity: mm.similarity ?? null,
|
|
127
|
+
};
|
|
128
|
+
});
|
|
129
|
+
return {
|
|
130
|
+
searchMode: obj.searchMode ?? null,
|
|
131
|
+
count: memories.length,
|
|
132
|
+
hits,
|
|
133
|
+
skipped: obj.skipped ?? false,
|
|
134
|
+
};
|
|
135
|
+
}
|
|
136
|
+
if (action === "search_notes") {
|
|
137
|
+
const results = Array.isArray(obj) ? obj : Array.isArray(obj.results) ? obj.results : [];
|
|
138
|
+
const hits = results.slice(0, 5).map((n) => {
|
|
139
|
+
const nn = n;
|
|
140
|
+
return {
|
|
141
|
+
id: nn.id,
|
|
142
|
+
title: typeof nn.title === "string" ? clip(nn.title, 120) : null,
|
|
143
|
+
score: nn.score ?? nn.similarity ?? null,
|
|
144
|
+
};
|
|
145
|
+
});
|
|
146
|
+
return { count: results.length, hits, skipped: obj.skipped ?? false };
|
|
147
|
+
}
|
|
148
|
+
if (action === "read_memories") {
|
|
149
|
+
const memories = Array.isArray(obj.memories) ? obj.memories : [];
|
|
150
|
+
const hits = memories.slice(0, 3).map((m) => {
|
|
151
|
+
const mm = m;
|
|
152
|
+
return {
|
|
153
|
+
id: mm.id,
|
|
154
|
+
memoryType: mm.memoryType,
|
|
155
|
+
summary: typeof mm.summary === "string" ? clip(mm.summary, 120) : null,
|
|
156
|
+
contentPreview: typeof mm.content === "string" ? clip(mm.content, 180) : null,
|
|
157
|
+
};
|
|
158
|
+
});
|
|
159
|
+
return { count: memories.length, hits };
|
|
160
|
+
}
|
|
161
|
+
if (action === "read_note") {
|
|
162
|
+
return {
|
|
163
|
+
id: obj.id ?? null,
|
|
164
|
+
title: typeof obj.title === "string" ? clip(obj.title, 120) : null,
|
|
165
|
+
contentPreview: typeof obj.content === "string" ? clip(obj.content, 220) : null,
|
|
166
|
+
};
|
|
167
|
+
}
|
|
168
|
+
if (action === "finish") {
|
|
169
|
+
return {
|
|
170
|
+
answerPreview: typeof obj.answer === "string" ? clip(obj.answer, 180) : null,
|
|
171
|
+
confidence: obj.confidence ?? null,
|
|
172
|
+
};
|
|
173
|
+
}
|
|
174
|
+
return undefined;
|
|
175
|
+
}
|
|
176
|
+
function summarizeStepForFallback(step) {
|
|
177
|
+
const raw = step.resultSnapshot
|
|
178
|
+
? JSON.stringify(step.resultSnapshot)
|
|
179
|
+
: step.result;
|
|
180
|
+
return clip(raw, state.tuning.summaryPerStepCharLimit);
|
|
181
|
+
}
|
|
182
|
+
function buildHeuristicFallbackAnswer() {
|
|
183
|
+
const latestSteps = state.steps.slice(-5);
|
|
184
|
+
const lines = [];
|
|
185
|
+
lines.push(`Goal: ${state.goal}`);
|
|
186
|
+
lines.push("Evidence collected (heuristic summary):");
|
|
187
|
+
for (const step of latestSteps) {
|
|
188
|
+
if (step.resultSnapshot) {
|
|
189
|
+
lines.push(`- ${step.action}: ${clip(JSON.stringify(step.resultSnapshot), 180)}`);
|
|
190
|
+
}
|
|
191
|
+
else if (step.result) {
|
|
192
|
+
lines.push(`- ${step.action}: ${clip(step.result, 180)}`);
|
|
193
|
+
}
|
|
194
|
+
}
|
|
195
|
+
if (state.retrieved.length > 0) {
|
|
196
|
+
lines.push("Top retrieved items:");
|
|
197
|
+
for (const item of state.retrieved.slice(-5)) {
|
|
198
|
+
lines.push(`- [${item.type}] ${item.summary}`);
|
|
199
|
+
}
|
|
200
|
+
}
|
|
201
|
+
lines.push("Run ended before explicit finish(); this is a best-effort summary from retrieved evidence.");
|
|
202
|
+
return lines.join("\n");
|
|
203
|
+
}
|
|
204
|
+
const result = await generateText({
|
|
205
|
+
model,
|
|
206
|
+
system: SYSTEM_PROMPT,
|
|
207
|
+
prompt: buildUserMessage(state),
|
|
208
|
+
tools,
|
|
209
|
+
stopWhen: stepCountIs(remainingSteps),
|
|
210
|
+
onStepFinish: (stepResult) => {
|
|
211
|
+
const stepStart = Date.now();
|
|
212
|
+
const { toolCalls, toolResults, usage } = stepResult;
|
|
213
|
+
if (!toolCalls || toolCalls.length === 0)
|
|
214
|
+
return;
|
|
215
|
+
for (let i = 0; i < toolCalls.length; i++) {
|
|
216
|
+
const call = toolCalls[i];
|
|
217
|
+
const toolResult = toolResults?.[i];
|
|
218
|
+
const parsedOutput = parseToolOutput(toolResult?.output);
|
|
219
|
+
const snapshot = buildResultSnapshot(call.toolName, parsedOutput);
|
|
220
|
+
if (call.toolName === "search_memories" && parsedOutput && typeof parsedOutput === "object") {
|
|
221
|
+
const mode = parsedOutput.searchMode;
|
|
222
|
+
if (mode === "fallback-recency") {
|
|
223
|
+
state.metrics.memoryFallbackCalls += 1;
|
|
224
|
+
}
|
|
225
|
+
}
|
|
226
|
+
const stepLog = {
|
|
227
|
+
action: call.toolName,
|
|
228
|
+
input: (call.input ?? {}),
|
|
229
|
+
result: typeof toolResult?.output === "string"
|
|
230
|
+
? clip(toolResult.output, state.tuning.stepResultCharLimit)
|
|
231
|
+
: clip(JSON.stringify(toolResult?.output), state.tuning.stepResultCharLimit),
|
|
232
|
+
resultSnapshot: snapshot,
|
|
233
|
+
tokensUsed: usage?.totalTokens ?? 0,
|
|
234
|
+
durationMs: Date.now() - stepStart,
|
|
235
|
+
};
|
|
236
|
+
state.steps.push(stepLog);
|
|
237
|
+
state.totalTokens += usage?.totalTokens ?? 0;
|
|
238
|
+
// Track retrieved items for context
|
|
239
|
+
if (call.toolName === "search_memories" || call.toolName === "search_notes") {
|
|
240
|
+
try {
|
|
241
|
+
const parsedObj = parsedOutput && typeof parsedOutput === "object"
|
|
242
|
+
? parsedOutput
|
|
243
|
+
: {};
|
|
244
|
+
const items = (parsedObj.memories ?? parsedObj.results ?? []);
|
|
245
|
+
for (const item of items.slice(0, 5)) {
|
|
246
|
+
state.retrieved.push({
|
|
247
|
+
type: call.toolName === "search_memories" ? "memory" : "note",
|
|
248
|
+
id: (item.id ?? item.noteId ?? "unknown"),
|
|
249
|
+
summary: (item.summary ?? item.title ?? item.content?.slice(0, 100) ?? ""),
|
|
250
|
+
});
|
|
251
|
+
}
|
|
252
|
+
}
|
|
253
|
+
catch {
|
|
254
|
+
// Parse failure — skip tracking
|
|
255
|
+
}
|
|
256
|
+
}
|
|
257
|
+
}
|
|
258
|
+
},
|
|
259
|
+
});
|
|
260
|
+
// Check if the LLM called finish()
|
|
261
|
+
let answer = "";
|
|
262
|
+
let confidence = 0;
|
|
263
|
+
let finishCalled = false;
|
|
264
|
+
for (const step of result.steps ?? []) {
|
|
265
|
+
for (const call of step.toolCalls ?? []) {
|
|
266
|
+
if (call.toolName === "finish") {
|
|
267
|
+
const input = call.input;
|
|
268
|
+
answer = input.answer;
|
|
269
|
+
confidence = input.confidence;
|
|
270
|
+
finishCalled = true;
|
|
271
|
+
}
|
|
272
|
+
}
|
|
273
|
+
}
|
|
274
|
+
const stepsRemaining = state.budget.maxSteps - state.steps.length;
|
|
275
|
+
if (!finishCalled &&
|
|
276
|
+
stepsRemaining <= state.tuning.forceFinalizeRemainingSteps &&
|
|
277
|
+
state.retrieved.length >= state.tuning.forceFinalizeMinEvidence) {
|
|
278
|
+
try {
|
|
279
|
+
const finalizeResult = await generateText({
|
|
280
|
+
model,
|
|
281
|
+
system: "You are finalizing a recursive retrieval run. Produce the best final answer from provided evidence and estimate confidence. Do not ask for more searches. Return strict JSON: {\"answer\":\"...\",\"confidence\":0.0}.",
|
|
282
|
+
prompt: [
|
|
283
|
+
`Goal: ${state.goal}`,
|
|
284
|
+
`\nEvidence summary:`,
|
|
285
|
+
...state.steps
|
|
286
|
+
.slice(-state.tuning.summaryMaxSteps)
|
|
287
|
+
.map((s) => `[${s.action}] ${summarizeStepForFallback(s)}`),
|
|
288
|
+
].join("\n"),
|
|
289
|
+
});
|
|
290
|
+
const parsedFinalize = parseFinalizePayload(finalizeResult.text ?? "");
|
|
291
|
+
answer = parsedFinalize?.answer ?? (finalizeResult.text || answer);
|
|
292
|
+
confidence = parsedFinalize?.confidence ?? Math.max(confidence, 0.45);
|
|
293
|
+
state.metrics.forcedFinalize = true;
|
|
294
|
+
state.totalTokens += finalizeResult.usage?.totalTokens ?? 0;
|
|
295
|
+
finishCalled = true;
|
|
296
|
+
state.finishReason = "controller_finalize";
|
|
297
|
+
}
|
|
298
|
+
catch {
|
|
299
|
+
// fallback handled below
|
|
300
|
+
}
|
|
301
|
+
}
|
|
302
|
+
// Fallback: if LLM never called finish(), synthesize answer from retrieved data
|
|
303
|
+
if (!finishCalled && state.steps.length > 0) {
|
|
304
|
+
if (disableLlmFallbackSummary) {
|
|
305
|
+
answer = buildHeuristicFallbackAnswer();
|
|
306
|
+
confidence = 0.25;
|
|
307
|
+
state.finishReason = state.budgetExhausted ? "budget_exhausted" : "controller_finalize";
|
|
308
|
+
state.confidence = confidence;
|
|
309
|
+
state.hypothesis = answer;
|
|
310
|
+
state.budgetExhausted = isBudgetExhausted(state);
|
|
311
|
+
if (state.budgetExhausted) {
|
|
312
|
+
state.finishReason = "budget_exhausted";
|
|
313
|
+
}
|
|
314
|
+
return { answer, confidence, state };
|
|
315
|
+
}
|
|
316
|
+
try {
|
|
317
|
+
const summaryResult = await generateText({
|
|
318
|
+
model,
|
|
319
|
+
system: "You are given retrieval results from a knowledge base search. Synthesize them into a clear, direct answer to the goal. Only state what the evidence supports. Be concise.",
|
|
320
|
+
prompt: [
|
|
321
|
+
`Goal: ${state.goal}`,
|
|
322
|
+
`\nRetrieval results:`,
|
|
323
|
+
...state.steps
|
|
324
|
+
.slice(-state.tuning.summaryMaxSteps)
|
|
325
|
+
.map((s) => `[${s.action}] ${summarizeStepForFallback(s)}`),
|
|
326
|
+
].join("\n"),
|
|
327
|
+
});
|
|
328
|
+
answer = summaryResult.text || "No answer could be generated.";
|
|
329
|
+
confidence = 0.3;
|
|
330
|
+
state.totalTokens += summaryResult.usage?.totalTokens ?? 0;
|
|
331
|
+
state.finishReason = state.budgetExhausted ? "budget_exhausted" : "controller_finalize";
|
|
332
|
+
}
|
|
333
|
+
catch {
|
|
334
|
+
// Last resort: use whatever text the model produced
|
|
335
|
+
answer = result.text || "Budget exhausted before the controller could finish.";
|
|
336
|
+
confidence = 0;
|
|
337
|
+
state.finishReason = state.budgetExhausted ? "budget_exhausted" : "controller_finalize";
|
|
338
|
+
}
|
|
339
|
+
}
|
|
340
|
+
else if (!finishCalled) {
|
|
341
|
+
answer = result.text || "No information found.";
|
|
342
|
+
confidence = 0;
|
|
343
|
+
state.finishReason = state.budgetExhausted ? "budget_exhausted" : "controller_finalize";
|
|
344
|
+
}
|
|
345
|
+
else if (!state.finishReason) {
|
|
346
|
+
state.finishReason = "tool_finish";
|
|
347
|
+
}
|
|
348
|
+
// Update total tokens from final usage
|
|
349
|
+
if (result.usage) {
|
|
350
|
+
const stepTokens = state.steps.reduce((sum, s) => sum + s.tokensUsed, 0);
|
|
351
|
+
const finalTotal = result.usage.totalTokens ?? 0;
|
|
352
|
+
if (finalTotal > stepTokens) {
|
|
353
|
+
state.totalTokens = finalTotal;
|
|
354
|
+
}
|
|
355
|
+
}
|
|
356
|
+
state.confidence = confidence;
|
|
357
|
+
state.hypothesis = answer;
|
|
358
|
+
state.budgetExhausted = isBudgetExhausted(state);
|
|
359
|
+
const finishedWithAnswer = finishCalled ||
|
|
360
|
+
state.finishReason === "tool_finish" ||
|
|
361
|
+
state.finishReason === "controller_finalize";
|
|
362
|
+
if (state.budgetExhausted && !finishedWithAnswer) {
|
|
363
|
+
state.finishReason = "budget_exhausted";
|
|
364
|
+
}
|
|
365
|
+
return { answer, confidence, state };
|
|
366
|
+
}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"execute.js","sourceRoot":"","sources":["../../src/rlm/execute.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,IAAI,CAAC;AAC/C,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAE9C,OAAO,EAAE,oBAAoB,EAAE,MAAM,cAAc,CAAC;AAGpD,MAAM,aAAa,GAAG;;;;;;;;;;;;;;;;4FAgBsE,CAAC;AAE7F,SAAS,gBAAgB,CAAC,KAAsB;IAC9C,MAAM,KAAK,GAAG,CAAC,SAAS,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;IAEtC,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QACrB,KAAK,CAAC,IAAI,CAAC,yBAAyB,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC/B,KAAK,CAAC,IAAI,CAAC,uBAAuB,KAAK,CAAC,SAAS,CAAC,MAAM,UAAU,CAAC,CAAC;QACpE,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;YAC9C,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAC5D,CAAC;IACH,CAAC;IAED,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,kBAAkB,KAAK,CAAC,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC9E,CAAC;IACD,MAAM,cAAc,GAAG,KAAK,CAAC,MAAM,CAAC,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC;IAClE,IAAI,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,2BAA2B,GAAG,CAAC,EAAE,CAAC;QACnE,KAAK,CAAC,IAAI,CACR,kCAAkC,cAAc,6DAA6D,CAC9G,CAAC;IACJ,CAAC;IAED,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,SAAS,CAAC;IAC7C,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,SAAS,GAAG,OAAO,CAAC;IACnD,IAAI,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,SAAS,GAAG,GAAG,EAAE,CAAC;QAC7C,KAAK,CAAC,IAAI,CAAC,gCAAgC,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACvG,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAsB;IAC/C,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC7D,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,SAAS,IAAI,KAAK,CAAC,MAAM,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IACxE,OAAO,KAAK,CAAC;AACf,CAAC;AAQD;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,GAAe,EACf,SAAoB,EACpB,KAAsB;IAEtB,MAAM,QAAQ,GAAG,YAAY,CAAC;QAC5B,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,OAAO,EAAE,SAAS,CAAC,OAAO;KAC3B,CAAC,CAAC;IACH,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;IAC/C,MAAM,KAAK,GAAG,oBAAoB,CAAC,GAAG,EAAE,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IAE9D,0CAA0C;IAC1C,MAAM,cAAc,GAAG,KAAK,CAAC,MAAM,CAAC,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC;IAElE,MAAM,yBAAyB,GAAG,KAAK,CAAC,MAAM,CAAC,yBAAyB,CAAC;IAEzE,SAAS,IAAI,CAAC,IAAY,EAAE,KAAa;QACvC,IAAI,IAAI,CAAC,MAAM,IAAI,KAAK;YAAE,OAAO,IAAI,CAAC;QACtC,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC;IACpC,CAAC;IAED,SAAS,eAAe,CAAC,KAAc;QACrC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC;YAAE,OAAO,SAAS,CAAC;QACvE,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC;IACzC,CAAC;IAED,SAAS,oBAAoB,CAAC,IAAY;QACxC,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAA+C,CAAC;YAC9E,OAAO;gBACL,MAAM,EAAE,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS;gBACrE,UAAU,EAAE,eAAe,CAAC,MAAM,CAAC,UAAU,CAAC;aAC/C,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,MAAM,UAAU,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;YACrC,MAAM,SAAS,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;YACxC,IAAI,UAAU,IAAI,CAAC,IAAI,SAAS,GAAG,UAAU,EAAE,CAAC;gBAC9C,IAAI,CAAC;oBACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,SAAS,GAAG,CAAC,CAAC,CAG9D,CAAC;oBACF,OAAO;wBACL,MAAM,EAAE,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS;wBACrE,UAAU,EAAE,eAAe,CAAC,MAAM,CAAC,UAAU,CAAC;qBAC/C,CAAC;gBACJ,CAAC;gBAAC,MAAM,CAAC;oBACP,OAAO,IAAI,CAAC;gBACd,CAAC;YACH,CAAC;YACD,OAAO,IAAI,CAAC;QACd,CAAC;IACH,CAAC;IAED,SAAS,eAAe,CAAC,GAAY;QACnC,IAAI,OAAO,GAAG,KAAK,QAAQ;YAAE,OAAO,GAAG,CAAC;QACxC,IAAI,CAAC;YACH,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACzB,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,GAAG,CAAC;QACb,CAAC;IACH,CAAC;IAED,SAAS,mBAAmB,CAC1B,MAAc,EACd,MAAe;QAEf,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ;YAAE,OAAO,SAAS,CAAC;QAC5D,MAAM,GAAG,GAAG,MAAiC,CAAC;QAE9C,IAAI,MAAM,KAAK,iBAAiB,EAAE,CAAC;YACjC,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;YACjE,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;gBAC1C,MAAM,EAAE,GAAG,CAA4B,CAAC;gBACxC,OAAO;oBACL,EAAE,EAAE,EAAE,CAAC,EAAE;oBACT,UAAU,EAAE,EAAE,CAAC,UAAU;oBACzB,OAAO,EAAE,OAAO,EAAE,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;oBACtE,UAAU,EAAE,EAAE,CAAC,UAAU,IAAI,IAAI;iBAClC,CAAC;YACJ,CAAC,CAAC,CAAC;YACH,OAAO;gBACL,UAAU,EAAE,GAAG,CAAC,UAAU,IAAI,IAAI;gBAClC,KAAK,EAAE,QAAQ,CAAC,MAAM;gBACtB,IAAI;gBACJ,OAAO,EAAE,GAAG,CAAC,OAAO,IAAI,KAAK;aAC9B,CAAC;QACJ,CAAC;QAED,IAAI,MAAM,KAAK,cAAc,EAAE,CAAC;YAC9B,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAE,GAAG,CAAC,OAAqB,CAAC,CAAC,CAAC,EAAE,CAAC;YACxG,MAAM,IAAI,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;gBACzC,MAAM,EAAE,GAAG,CAA4B,CAAC;gBACxC,OAAO;oBACL,EAAE,EAAE,EAAE,CAAC,EAAE;oBACT,KAAK,EAAE,OAAO,EAAE,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;oBAChE,KAAK,EAAE,EAAE,CAAC,KAAK,IAAI,EAAE,CAAC,UAAU,IAAI,IAAI;iBACzC,CAAC;YACJ,CAAC,CAAC,CAAC;YACH,OAAO,EAAE,KAAK,EAAE,OAAO,CAAC,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,IAAI,KAAK,EAAE,CAAC;QACxE,CAAC;QAED,IAAI,MAAM,KAAK,eAAe,EAAE,CAAC;YAC/B,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;YACjE,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;gBAC1C,MAAM,EAAE,GAAG,CAA4B,CAAC;gBACxC,OAAO;oBACL,EAAE,EAAE,EAAE,CAAC,EAAE;oBACT,UAAU,EAAE,EAAE,CAAC,UAAU;oBACzB,OAAO,EAAE,OAAO,EAAE,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;oBACtE,cAAc,EACZ,OAAO,EAAE,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;iBAChE,CAAC;YACJ,CAAC,CAAC,CAAC;YACH,OAAO,EAAE,KAAK,EAAE,QAAQ,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC;QAC1C,CAAC;QAED,IAAI,MAAM,KAAK,WAAW,EAAE,CAAC;YAC3B,OAAO;gBACL,EAAE,EAAE,GAAG,CAAC,EAAE,IAAI,IAAI;gBAClB,KAAK,EAAE,OAAO,GAAG,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;gBAClE,cAAc,EACZ,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;aAClE,CAAC;QACJ,CAAC;QAED,IAAI,MAAM,KAAK,QAAQ,EAAE,CAAC;YACxB,OAAO;gBACL,aAAa,EACX,OAAO,GAAG,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;gBAC/D,UAAU,EAAE,GAAG,CAAC,UAAU,IAAI,IAAI;aACnC,CAAC;QACJ,CAAC;QAED,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,SAAS,wBAAwB,CAAC,IAAa;QAC7C,MAAM,GAAG,GAAG,IAAI,CAAC,cAAc;YAC7B,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,cAAc,CAAC;YACrC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC;QAChB,OAAO,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,CAAC,CAAC;IACzD,CAAC;IAED,SAAS,4BAA4B;QACnC,MAAM,WAAW,GAAG,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;QAC1C,MAAM,KAAK,GAAa,EAAE,CAAC;QAE3B,KAAK,CAAC,IAAI,CAAC,SAAS,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;QAClC,KAAK,CAAC,IAAI,CAAC,yCAAyC,CAAC,CAAC;QAEtD,KAAK,MAAM,IAAI,IAAI,WAAW,EAAE,CAAC;YAC/B,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;gBACxB,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,MAAM,KAAK,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,cAAc,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;YACpF,CAAC;iBAAM,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;gBACvB,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,MAAM,KAAK,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;YAC5D,CAAC;QACH,CAAC;QAED,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;YACnC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC7C,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;YACjD,CAAC;QACH,CAAC;QAED,KAAK,CAAC,IAAI,CACR,4FAA4F,CAC7F,CAAC;QACF,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAED,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC;QAChC,KAAK;QACL,MAAM,EAAE,aAAa;QACrB,MAAM,EAAE,gBAAgB,CAAC,KAAK,CAAC;QAC/B,KAAK;QACL,QAAQ,EAAE,WAAW,CAAC,cAAc,CAAC;QACrC,YAAY,EAAE,CAAC,UAAU,EAAE,EAAE;YAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,GAAG,UAAU,CAAC;YACrD,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;gBAAE,OAAO;YAEjD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC1C,MAAM,IAAI,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;gBAC1B,MAAM,UAAU,GAAG,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC;gBACpC,MAAM,YAAY,GAAG,eAAe,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;gBACzD,MAAM,QAAQ,GAAG,mBAAmB,CAAC,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;gBAClE,IAAI,IAAI,CAAC,QAAQ,KAAK,iBAAiB,IAAI,YAAY,IAAI,OAAO,YAAY,KAAK,QAAQ,EAAE,CAAC;oBAC5F,MAAM,IAAI,GAAI,YAAwC,CAAC,UAAU,CAAC;oBAClE,IAAI,IAAI,KAAK,kBAAkB,EAAE,CAAC;wBAChC,KAAK,CAAC,OAAO,CAAC,mBAAmB,IAAI,CAAC,CAAC;oBACzC,CAAC;gBACH,CAAC;gBAED,MAAM,OAAO,GAAY;oBACvB,MAAM,EAAE,IAAI,CAAC,QAAQ;oBACrB,KAAK,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAA4B;oBACpD,MAAM,EACJ,OAAO,UAAU,EAAE,MAAM,KAAK,QAAQ;wBACpC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,mBAAmB,CAAC;wBAC3D,CAAC,CAAC,IAAI,CACF,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE,MAAM,CAAC,EAClC,KAAK,CAAC,MAAM,CAAC,mBAAmB,CACjC;oBACP,cAAc,EAAE,QAAQ;oBACxB,UAAU,EAAE,KAAK,EAAE,WAAW,IAAI,CAAC;oBACnC,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;iBACnC,CAAC;gBACF,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;gBAC1B,KAAK,CAAC,WAAW,IAAI,KAAK,EAAE,WAAW,IAAI,CAAC,CAAC;gBAE7C,oCAAoC;gBACpC,IAAI,IAAI,CAAC,QAAQ,KAAK,iBAAiB,IAAI,IAAI,CAAC,QAAQ,KAAK,cAAc,EAAE,CAAC;oBAC5E,IAAI,CAAC;wBACH,MAAM,SAAS,GACb,YAAY,IAAI,OAAO,YAAY,KAAK,QAAQ;4BAC9C,CAAC,CAAE,YAAwC;4BAC3C,CAAC,CAAC,EAAE,CAAC;wBACT,MAAM,KAAK,GAAG,CAAC,SAAS,CAAC,QAAQ,IAAI,SAAS,CAAC,OAAO,IAAI,EAAE,CAE3D,CAAC;wBACF,KAAK,MAAM,IAAI,IAAK,KAAwC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;4BACzE,KAAK,CAAC,SAAS,CAAC,IAAI,CAAC;gCACnB,IAAI,EAAE,IAAI,CAAC,QAAQ,KAAK,iBAAiB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM;gCAC7D,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,IAAI,IAAI,CAAC,MAAM,IAAI,SAAS,CAAW;gCACnD,OAAO,EAAE,CAAC,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,KAAK,IAAK,IAAI,CAAC,OAAkB,EAAE,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,EAAE,CAAW;6BACjG,CAAC,CAAC;wBACL,CAAC;oBACH,CAAC;oBAAC,MAAM,CAAC;wBACP,gCAAgC;oBAClC,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC,CAAC;IAEH,mCAAmC;IACnC,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,YAAY,GAAG,KAAK,CAAC;IAEzB,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,IAAI,EAAE,EAAE,CAAC;QACtC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,SAAS,IAAI,EAAE,EAAE,CAAC;YACxC,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;gBAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,KAA+C,CAAC;gBACnE,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;gBACtB,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC;gBAC9B,YAAY,GAAG,IAAI,CAAC;YACtB,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,KAAK,CAAC,MAAM,CAAC,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC;IAClE,IACE,CAAC,YAAY;QACb,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,2BAA2B;QAC1D,KAAK,CAAC,SAAS,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,wBAAwB,EAC/D,CAAC;QACD,IAAI,CAAC;YACH,MAAM,cAAc,GAAG,MAAM,YAAY,CAAC;gBACxC,KAAK;gBACL,MAAM,EACJ,wNAAwN;gBAC1N,MAAM,EAAE;oBACN,SAAS,KAAK,CAAC,IAAI,EAAE;oBACrB,qBAAqB;oBACrB,GAAG,KAAK,CAAC,KAAK;yBACX,KAAK,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC;yBACpC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,KAAK,wBAAwB,CAAC,CAAC,CAAC,EAAE,CAAC;iBAC9D,CAAC,IAAI,CAAC,IAAI,CAAC;aACb,CAAC,CAAC;YACH,MAAM,cAAc,GAAG,oBAAoB,CAAC,cAAc,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;YACvE,MAAM,GAAG,cAAc,EAAE,MAAM,IAAI,CAAC,cAAc,CAAC,IAAI,IAAI,MAAM,CAAC,CAAC;YACnE,UAAU,GAAG,cAAc,EAAE,UAAU,IAAI,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC;YACtE,KAAK,CAAC,OAAO,CAAC,cAAc,GAAG,IAAI,CAAC;YACpC,KAAK,CAAC,WAAW,IAAI,cAAc,CAAC,KAAK,EAAE,WAAW,IAAI,CAAC,CAAC;YAC5D,YAAY,GAAG,IAAI,CAAC;YACpB,KAAK,CAAC,YAAY,GAAG,qBAAqB,CAAC;QAC7C,CAAC;QAAC,MAAM,CAAC;YACP,yBAAyB;QAC3B,CAAC;IACH,CAAC;IAED,gFAAgF;IAChF,IAAI,CAAC,YAAY,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5C,IAAI,yBAAyB,EAAE,CAAC;YAC9B,MAAM,GAAG,4BAA4B,EAAE,CAAC;YACxC,UAAU,GAAG,IAAI,CAAC;YAClB,KAAK,CAAC,YAAY,GAAG,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,qBAAqB,CAAC;YACxF,KAAK,CAAC,UAAU,GAAG,UAAU,CAAC;YAC9B,KAAK,CAAC,UAAU,GAAG,MAAM,CAAC;YAC1B,KAAK,CAAC,eAAe,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;YACjD,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC;gBAC1B,KAAK,CAAC,YAAY,GAAG,kBAAkB,CAAC;YAC1C,CAAC;YACD,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC;QACvC,CAAC;QACD,IAAI,CAAC;YACH,MAAM,aAAa,GAAG,MAAM,YAAY,CAAC;gBACvC,KAAK;gBACL,MAAM,EAAE,2KAA2K;gBACnL,MAAM,EAAE;oBACN,SAAS,KAAK,CAAC,IAAI,EAAE;oBACrB,sBAAsB;oBACtB,GAAG,KAAK,CAAC,KAAK;yBACX,KAAK,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC;yBACpC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,KAAK,wBAAwB,CAAC,CAAC,CAAC,EAAE,CAAC;iBAC9D,CAAC,IAAI,CAAC,IAAI,CAAC;aACb,CAAC,CAAC;YACH,MAAM,GAAG,aAAa,CAAC,IAAI,IAAI,+BAA+B,CAAC;YAC/D,UAAU,GAAG,GAAG,CAAC;YACjB,KAAK,CAAC,WAAW,IAAI,aAAa,CAAC,KAAK,EAAE,WAAW,IAAI,CAAC,CAAC;YAC3D,KAAK,CAAC,YAAY,GAAG,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,qBAAqB,CAAC;QAC1F,CAAC;QAAC,MAAM,CAAC;YACP,oDAAoD;YACpD,MAAM,GAAG,MAAM,CAAC,IAAI,IAAI,sDAAsD,CAAC;YAC/E,UAAU,GAAG,CAAC,CAAC;YACf,KAAK,CAAC,YAAY,GAAG,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,qBAAqB,CAAC;QAC1F,CAAC;IACH,CAAC;SAAM,IAAI,CAAC,YAAY,EAAE,CAAC;QACzB,MAAM,GAAG,MAAM,CAAC,IAAI,IAAI,uBAAuB,CAAC;QAChD,UAAU,GAAG,CAAC,CAAC;QACf,KAAK,CAAC,YAAY,GAAG,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,qBAAqB,CAAC;IAC1F,CAAC;SAAM,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,CAAC;QAC/B,KAAK,CAAC,YAAY,GAAG,aAAa,CAAC;IACrC,CAAC;IAED,uCAAuC;IACvC,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QACjB,MAAM,UAAU,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC;QACzE,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC;QACjD,IAAI,UAAU,GAAG,UAAU,EAAE,CAAC;YAC5B,KAAK,CAAC,WAAW,GAAG,UAAU,CAAC;QACjC,CAAC;IACH,CAAC;IAED,KAAK,CAAC,UAAU,GAAG,UAAU,CAAC;IAC9B,KAAK,CAAC,UAAU,GAAG,MAAM,CAAC;IAC1B,KAAK,CAAC,eAAe,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;IACjD,MAAM,kBAAkB,GACtB,YAAY;QACZ,KAAK,CAAC,YAAY,KAAK,aAAa;QACpC,KAAK,CAAC,YAAY,KAAK,qBAAqB,CAAC;IAC/C,IAAI,KAAK,CAAC,eAAe,IAAI,CAAC,kBAAkB,EAAE,CAAC;QACjD,KAAK,CAAC,YAAY,GAAG,kBAAkB,CAAC;IAC1C,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC;AACvC,CAAC"}
|
|
@@ -0,0 +1,6 @@
|
|
|
1
|
+
import type { McpContext, LlmConfig } from "../auth.js";
|
|
2
|
+
import { type ControllerInput, type ControllerResult } from "./types.js";
|
|
3
|
+
/**
|
|
4
|
+
* Run the RLM controller: recursive retrieval, verification, and writeback.
|
|
5
|
+
*/
|
|
6
|
+
export declare function runController(ctx: McpContext, llmConfig: LlmConfig, input: ControllerInput): Promise<ControllerResult>;
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/rlm/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAKxD,OAAO,EAAkB,KAAK,eAAe,EAAE,KAAK,gBAAgB,EAAwB,MAAM,YAAY,CAAC;AAkB/G;;GAEG;AACH,wBAAsB,aAAa,CACjC,GAAG,EAAE,UAAU,EACf,SAAS,EAAE,SAAS,EACpB,KAAK,EAAE,eAAe,GACrB,OAAO,CAAC,gBAAgB,CAAC,CAgJ3B"}
|