Mnemon -- execution caching and system learning for AI agents. Zero code changes. One import line. Gets smarter with every run. Saves tokens on every cache hit. Usage: import mnemo

Fetches usage statistics and project information for an organization’s billing overview page.

Anonymous opt-out telemetry — PostHog. Disable by setting MNEMON_NO_TELEMETRY=1 in the environment. What is sent: framework, cache_level, tokens_saved, latency_ms, mnemon_version,

Audit log — tracks every outbound request made during autopilot sessions. Append-only JSONL file at hunt-memory/audit.jsonl. Used for post-session review and scope compliance verif

Canonical NATS subject and stream name helpers. ! ! The cloud enrollment response provides a tenant-scoped `subject_prefix`. ! All adaptive-SDR NATS channels must derive from that

Sends embedding operation metrics to Langfuse, constructing tags and metadata for observability of AI embeddings.

Model cost accounting and per-agent budgets (Milestone C). Every model call records its token usage and dollar cost on the `model.call` audit event. This module is the math and the

Initializes OpenPanel SDK for server‑side tracking with API credentials from env. Reuse for any server analytics integration.

Builds causal impact data for changed policy events by extracting their causal context from a causal graph, identifying breakage drivers through root node analysis.

Adaptive compression sizing via information saturation detection. ! ! Direct port of `headroom/transforms/adaptive_sizer.py`. Used by ! `smart_crusher`'s array crushers to decide *

Per-run token/cost accounting, audit log, and risk-level/human-in-the-loop compliance config declared in agent.yaml. Worth extracting given how expensive untracked multi-agent sess

Exposes derived Svelte stores for the current project, its region, onboarding status, and regional console variables.

SemanticDedup — near-duplicate content block elimination FusionStage. Detects and eliminates repeated content blocks within a single text using 3-word shingle fingerprinting (no ex

Computes weighted shortest exploitation paths in a Neo4j attack graph, integrating edge difficulty, severity, and PoC validation to plan multi-step attack chains.

Root layout sets HTML/body, Geist fonts, metadata, analytics.

Mnemon Telemetry Structured metrics and observability. Collects what every component already produces and puts it in one place. Metrics tracked: - EME: hit rates by level, tokens s

Modal dialog for editing DNS records with form inputs, validation, and analytics tracking; used wherever DNS record modifications are needed.

Wraps Plausible and generic analytics, exposing trackEvent, trackError, and page view tracking for Svelte applications.

Collects, sanitizes, and writes OpenTelemetry trace data, providing span creation and event logging utilities.

Computes weighted rating for reputation scoring in trust aggregation.

Formats duration seconds into a human-readable h/m/s string for performance uptime display.

Pure functions that compute mean, stddev and detect statistical anomalies from daily‑count buckets for the find_anomalies operation.

Compute weighted aggregation of component ratings into overall score.

Loads built-in source definitions from bundled configuration into registry.

Reporting the call back to call-api. Delivery is **at-least-once, keyed by job_id** (BRIEF §9 phase-2 trap 4). Two properties matter and neither is negotiable: - A lost POST must n

Reusable button component for forms, supporting submit actions, loading states, analytics events, and link behavior with theme‑aware styling.

Ranking-quality metrics: nDCG@k, MRR@k, Recall@k. Pure-Python, dependency-free. Each metric takes ``predicted`` (the ranked list of IDs returned by the system under test) and ``rel

Computes of 10-5 to,

Data model representing a health check result with status, latency, and timestamp, used to track service health over time.

Paginates daily stats for leaderboards; retrieves trend entries with cursor, keyed by date.

Multilayer mounts nested Kubernetes volumes into container storage layers.

Evaluates agent evidence against eight CVE‑Bench win conditions, producing verdicts with pass/fail, met conditions and timing.

Maps ExploitBench V8 capability flags to tier levels and computes weighted scores matching the official leaderboard methodology.

Benchmarks a custom evaluator against labelled datasets, producing predictions and performance metrics.

Generates project status reports, risk analysis, sprint summaries, and effort estimates using read‑only project data.

Captures, flushes, and manages telemetry events via PostHog, handling disabled mode and error reporting.

Defines constant event names and a helper to track analytics actions across the application.

Extracts node IDs from causal paths to generate policy enforcement promotion suggestions based on historical causal impact analysis.

Displays real‑time analytics of sessions, token usage, and events using heatmaps, trend lines, and interactive charts refreshed via WebSocket.

Classifies an endpoint URL into categories like localhost, private_network, or remote based on hostname and IP patterns, used for telemetry filtering.

Singleton async client for LangFuse telemetry, lazily initialized and container‑aware for observability integration.

Computes a percentile from a list of floats using linear interpolation, returning 0 for empty input. Useful for summarizing benchmark result distributions.

L6: Observation Compression (AGGRESSIVE) Inspired by claw-compactor's 97% compression on tool results. Tool call results (especially large ones) are summarized to key info only. Th

Extracts printable strings from binary blobs and tags them into categories like URL, IP, crypto, version, aiding automated knowledge‑graph enrichment.

Executes primary fixture‑based and secondary replay‑based quality gates for expert retrieval, outputting pass/fail verdicts and optional JSON reports.

Computes structured, severity‑rated diagnostics for kanban tasks, providing actionable recovery hints for UI and CLI.

Async service that records LLM step metadata to the database and retrieves step records for analytics.

Computes per‑source freshness, staleness and activity metrics for dashboard and health checks, optimizing queries for large brain datasets.

Shows users how many tokens RTK has saved them over time.

Builds OpenSearch queries for analytics endpoints, converting filter structures into JSON values for search operations.