cost-attribution

Self-contained DeepSeek Harness (DSH) plugin for Provider/Auth login, model switching, image fallback, token/cost analytics, and same-port Web restart. Useful? A star helps.

Databricks App for cross-workspace AI stack governance. Unified view of agents, serving endpoints, knowledge bases, and AI Gateway usage, backed by system tables + Lakebase with auto-refreshing discovery workflows.

Vidai — the AI control plane. Govern, cost and secure every enterprise LLM request, from within your own network. Self-install Docker releases.

OTel-native typed primitives for LLM cost attribution and telemetry — published on PyPI.

DSH (DeepSeek Harness) plugin: per-session cost line + cost attribution report, priced by llm-pricing

PyPI-distributable LLM control plane: gateway choke point, cost attribution, OTel instrumentation, and offline reporting as an inspectable engineering artifact.

Per-subagent cost attribution for Claude Code. Reads local JSONL session logs, computes shadow cost at marginal Anthropic API rates with proper TTL pricing.

OpenTelemetry-native token/cost attribution for LLM agents, with budgets that degrade before they deny

Infer ownership of untagged AWS resources from CloudTrail behaviour and write tags back — the open-source FinOps input layer. Methodology peer-validated on Microsoft Azure 2.6M-VM dataset.

Open-source LLM API cost workbench with a public source-linked Pricing Feed v1, browser-local bill audit, workflow ledger, and fixed-price implementation scopes.

Attribute Claude Code and Codex token costs to directories, packages, teams, and features.

FinOps cost attribution for AI code agents: maps Kiro, Cursor, and Claude Code spend to git commits to reveal per-task cost, waste patterns, and agent ROI signals.

Per-PR preview environments for Kubernetes with TTL cleanup and actual cost-on-PR attribution (OpenCost)

Pre-dispatch policy evaluation and cost attribution for LLM inference, built on llmscope.

AI FinOps Engine - GPU/LLM cost attribution for Kubernetes

The visibility gap LiteLLM doesn't close: attribute LLM spend per RAG pipeline stage (retrieval / reranking / generation / evaluation). OpenAI-compatible, Ollama-backed, <8ms overhead.

Cost attribution for LLM serving with a prefix cache. The only bill that sums to what the server spent charges the tenant that arrived first, in 360 of 360 family replays. Exact Shapley in one pass over the prefix trie. Exit 2 when a scheme is not a division of anything. Every README figure re-measured by CI.

Attribute cloud database spend to the teams, apps and features that caused it — joins pg_stat_statements deltas with billing exports to produce a defensible chargeback report.

A token-efficiency meter + playbook for agentic coding: measure where your tokens and dollars go per stage, catch silently-degraded runs, and cut cost without losing capability. Stdlib-only.

Two-stage LLM request router — classifies complexity with embedding similarity + DeBERTa zero-shot, routes to the cheapest capable model, and improves from its own mistakes. Fully local with Ollama.