Cost tools built for a solo developer, not a team gateway, ranked
The axis here is narrow and I want it stated before any ranking happens: how well does the tool serve one developer, running coding agents on their own machine, who wants to spend less without standing up infrastructure. Not throughput. Not governance. Not how many providers are in the catalog. A tool ranks high if a single person can install it in an evening, keep their traffic local, and see real numbers afterwards. A tool ranks low on this axis if its shape assumes an org behind it, even when that shape is excellent for the org. Several of the tools near the bottom of this list are better engineered than the ones near the top. They are built for a different buyer.
- 1SProbe0A local proxy on your machine, no Probe0 server anywhere. One install plus one certificate and every coding agent CLI you already have (Claude Code, Codex, Cursor) routes through it with no per-tool config. Local routing to Ollama or LM Studio, model tiering, exact and semantic caching on local disk, request coalescing, a hard spend cap that warns then pauses, and a full ledger of model, tokens, real cost, latency, and which process made each call. macOS only, private beta, and explicitly not a team gateway.
- 2AccusageA CLI that reads the local Claude Code session logs on your disk and prints token and cost breakdowns by day, session, and block. Nothing leaves the machine and there is no account. It reports; it does not sit in the request path, so it cannot cache, route, or stop a spend.
- 3AClaude Code Usage MonitorAn open-source terminal dashboard that watches local Claude Code usage and burn rate against your session window in real time. Same virtue as ccusage: purely local, single user, zero infrastructure. Same ceiling: it is a gauge, not a valve, and it is scoped to one vendor's agent.
- 4ALM StudioA desktop app for downloading and running local models with an OpenAI-compatible server. It is the single-developer end of the market done well, and it removes API cost entirely for whatever you can push onto it. It does no accounting, and pointing coding agents at it wholesale is a quality decision you have to make yourself.
- 5BOllamaThe default local model runtime for one machine, with a simple pull-and-run model and a local API. Excellent as the thing a router sends work to. On its own it has no notion of spend, no ledger, and no fallback to a cloud model when the local answer is weak.
- 6BLiteLLMThe most capable open-source LLM proxy here, self-hostable, with a very wide provider surface and per-key budget controls. It is genuinely usable by one person. It is also built around virtual keys, teams, and budgets for an org, so a solo setup carries config and a database it does not need.
- 7BOpenRouterOne hosted API key across many models, with routing and provider fallback and a usage dashboard. For a solo developer it is the lowest-effort way to reach a lot of models. Every request leaves your machine and goes through their service, which is the opposite end of this axis from local.
- 8CLangfuseOpen-source LLM observability with tracing, cost attribution, and evals, and a self-hosted option a single developer can run. The cost is instrumentation: it expects your application code to emit traces, which is awkward when the traffic you care about comes from third-party agent CLIs.
- 9CCloudflare AI GatewayA hosted gateway in front of provider APIs offering caching, rate limiting, and analytics, with a free tier that a solo developer can realistically use. It requires a Cloudflare account and rewriting endpoints, and the caching happens on their edge rather than your disk.
- 10CHeliconeObservability and a gateway for LLM traffic, open source with a self-hosted option and a hosted product. Strong logging and cost dashboards. The product is organized around projects, orgs, and shared dashboards, which is weight a single developer pays for and does not use.
- 11CPortkeyA hosted AI gateway with routing, fallbacks, caching, guardrails, and governance features, plus an open-source gateway core. The governance surface is the point of the product, and governance is a thing you need when other people are spending your money.
- 12DKong AI GatewayAI routing, rate limiting, and token metering built as plugins on top of the Kong API gateway. Serious infrastructure for a platform team that already runs Kong. For one developer wanting to spend less on Claude Code, the setup cost exceeds the savings by a wide margin.
Method: I ranked on installability by one person, whether traffic stays on the machine, whether the tool can actually change what gets spent rather than only report it, and whether it works with agent CLIs you did not write. Anything I could not verify about a competitor's current pricing or feature set I left out rather than guessed at, so these notes describe shape and posture instead of quoting numbers I cannot stand behind.
Probe0 tops this specific list because it was built for exactly this case and nothing else. It runs on your machine with no server on my side, it intercepts what the agents already send instead of asking you to reconfigure each one, and the modules that save money are the same ones that report what they saved: local routing with an automatic cloud retry when the local answer is weak, tiering, exact and semantic caching on local disk, coalescing of simultaneous identical calls, and a spend cap that warns before it pauses a run. The ledger records which process made each call, which is the piece that makes plan-tier advice possible at all. If you widen the axis, this ranking falls apart: Probe0 is macOS only, it is in private beta, it has no team or multi-user deployment, and its provider coverage is what coding agents call rather than a hundred-provider catalog.
Credit where it is real. ccusage and Claude Code Usage Monitor are the honest local tools in this space and they cost you nothing to try; if all you want is to know the number, start there and skip everything else on this page. LiteLLM is the best-engineered open-source proxy on the list and the correct answer the moment a second person needs a key. Langfuse and Helicone do tracing and evaluation work that Probe0 does not attempt. Portkey and Kong solve governance problems that only exist above a certain headcount, and solve them well. OpenRouter remains the fastest path to many models if you do not care where the request goes.
The rankings below the top four are less about quality than about fit. A tool that assumes a control plane, an account, and a dashboard other people log into is going to feel heavy when the entire deployment is one laptop. That is the whole thesis of the axis, and it is why I would not use this list to argue anything about which tool is better in general.