agentmoney

Quick start · Website · How it works · AGENTS.md · llms.txt · Skills · Discussions · All projects

Know what your AI costs: pricing and spend for Claude, GPT and Gemini.

AgentMoney is an agent employee, a packaged specialist agent with its own skills, memory and tests. It owns one job: observability. Its persona is Ledger. Coding agents use it through its skills, its MCP server and its command line. A scheduled job runs its evals and proposes fixes. A person makes every change.

AgentMoney is part of an agent-native play for agent-native systems that can eventually be recursively self-improving and evolving, operated by agent employees.

agentmoney tells you what your AI costs. It prices one call, a session, or every call your app makes, from a cited rate table of 28 Claude, OpenAI and Gemini models, and it reads the Anthropic Usage API for what you actually spent. It is for anyone who runs agents or LLM apps: you run Claude Code overnight, the bill is $247, and you want to know which session burned it, which model, and whether prompt caching saved anything.

Use it from any coding agent

Claude Code: the plugin brings the skills, sub-agents, hooks and MCP server.

/plugin marketplace add naman10parikh/agentmoney
/plugin install agentmoney@agentmoney

Its MCP server, Codex, Cursor, Gemini CLI, one skill and the command line

Its MCP server, in any MCP client:

claude mcp add --transport stdio agentmoney -- npx -y -p github:naman10parikh/agentmoney agentmoney-mcp
codex mcp add agentmoney -- npx -y -p github:naman10parikh/agentmoney agentmoney-mcp

Cursor (.cursor/mcp.json), Gemini CLI (.gemini/settings.json) and most other clients take the same command as JSON:

{
  "mcpServers": {
    "agentmoney": {
      "command": "npx",
      "args": [
        "-y",
        "-p",
        "github:naman10parikh/agentmoney",
        "agentmoney-mcp"
      ]
    }
  }
}

Gemini CLI: the extension brings the context and the MCP server.

gemini extensions install https://github.com/naman10parikh/agentmoney

One skill (3 in all, in skills/), in any agent that reads Agent Skills:

gemini skills install https://github.com/naman10parikh/agentmoney --path skills/cost-investigation
git clone --depth 1 https://github.com/naman10parikh/agentmoney /tmp/agentmoney && cp -r /tmp/agentmoney/skills/cost-investigation ~/.claude/skills/

The command line:

npx -y -p github:naman10parikh/agentmoney agentmoney --help

Codex, Cursor, OpenCode, GitHub Copilot and other agents read AGENTS.md; Claude Code reads CLAUDE.md and Gemini CLI reads GEMINI.md, which import it. llms.txt maps every surface for agents that read first.

Quick start

You need Node 18 or newer (sandbox-run needs Node 20). calc, estimate, models and demo need nothing else and never touch the network.

# Run it straight from GitHub, nothing installed: price 8 sample calls and see the breakdown
npx -y -p github:naman10parikh/agentmoney agentmoney demo

# Or keep a copy: clone, build, and put the commands on your PATH
git clone https://github.com/naman10parikh/agentmoney && cd agentmoney
npm install && npm install -g .

Price a call or a session:

agentmoney calc -m claude-sonnet-4-6 -i 5000 -o 1000                    # one call: $0.030000
agentmoney calc -m claude-sonnet-4-6 -i 5000 -o 1000 --cache-read 3000  # with prompt caching: $0.030900
agentmoney estimate -m claude-sonnet-4-6 -c 50 --avg-input 3000 --avg-output 800   # 50 calls: $1.0500
agentmoney models                                                       # every model and its rates

Track the calls your app really makes:

agentmoney proxy --port 8999 --alert 5.00        # warn when a session passes $5
ANTHROPIC_BASE_URL=http://localhost:8999 your-app
# live dashboard: http://localhost:8999/dashboard   cost JSON: http://localhost:8999/costs

Read what Anthropic recorded. This needs an Admin API key (it starts with sk-ant-admin); workspace API keys cannot read usage.

export ANTHROPIC_ADMIN_KEY=...
agentmoney usage                    # the last 7 days, today included
agentmoney usage --days 30 --json

Example output

agentmoney demo prices 8 sample calls with a $0.05 session alert, so it also shows the alert firing:

agentmoney demo

Simulating 8 API calls...

.[agentmoney] WARNING: Session cost $0.0549 exceeded $0.05 threshold
.[agentmoney] WARNING: Session cost $0.0584 exceeded $0.05 threshold
...
agentmoney summary
──────────────────────────────────────────────────

  Total cost:      $0.2812
  Total calls:     8
  Sessions:        1

  By provider:
    anthropic       $0.2702  (7 calls)
    openai          $0.0110  (1 calls)

  By model:
    claude-sonnet-4-6                 $0.1998  (4 calls, 35.8K tokens)
    claude-haiku-4-5                $0.005400  (2 calls, 3.0K tokens)
    claude-opus-4-6                   $0.0650  (1 calls, 5.8K tokens)
    gpt-4o                            $0.0110  (1 calls, 2.6K tokens)

  Session demo: $0.2812 (8 calls)

Reports saved: .agentmoney/report-YYYY-MM-DD.json, .agentmoney/report-YYYY-MM-DD.md

How it works

Mode	What happens
`calc`, `estimate`, `models`	Pure math: model and tokens in, cost out, from the rate table in `src/pricing.ts`. No network.
`proxy`	An Express server forwards every request to the target, reads the `usage` block of each JSON response or the usage events of a streamed one (Anthropic's or OpenAI's, told apart by their fields), prices it, logs it and serves a dashboard.
`usage`	Reads `GET /v1/organizations/usage_report/messages` (daily, grouped by model, every page) and prices the tokens with the same engine. A one-hour cache write costs twice the input rate; web search and code execution charges are not included.
`sandbox-run`	Prices a call log you did not write inside an E2B microVM, with a program generated from the same rate table, and checks the total against the host. Needs `E2B_API_KEY`.
MCP `cost_query`	Any coding agent prices a call by asking the `agentmoney-mcp` server.

The proxy

Your app → agentmoney proxy (localhost:8999) → api.anthropic.com
                ↓
          cost tracking per model, provider and session
          web dashboard (Chart.js)
          .agentmoney/calls.jsonl
          daily reports

Path	What it returns
`GET /dashboard`	the web dashboard: cost per session per day, the hourly trend and the split by model. The charts load Chart.js from the jsDelivr CDN, so they need the internet. Offline, the page says so and still shows the totals and the calls.
`GET /health`	proxy status, total cost and call count (JSON)
`GET /costs`	the full breakdown by model, provider and session (JSON)
`* /*`	every other request, forwarded to the target

The proxy passes a streamed reply on to your app event by event, as it arrives. It prices the reply from its events once the stream ends: Anthropic's message_start and message_delta, or OpenAI's last chunk. An OpenAI stream carries usage only when the request sets stream_options.include_usage; without it, the call is not priced.

--alert <usd> prints a warning each time a session's cost passes the threshold. Stop the proxy with Ctrl+C and it prints a summary and writes the day's reports.

Reports

agentmoney writes to .agentmoney/ in the folder it runs in:

calls.jsonl: every call the proxy saw, one JSON line each (model, tokens, cost, duration, endpoint, session)
report-YYYY-MM-DD.json: the day's cost report as data
report-YYYY-MM-DD.md: the same report in Markdown, ready to share

Commands

Command	What it does
`agentmoney`	show the commands and a quick start
`agentmoney demo`	price 8 sample calls and write a report
`agentmoney calc`	price one call
`agentmoney estimate`	price a session of similar calls
`agentmoney models`	list every model and its rates
`agentmoney proxy`	start the cost-tracking proxy and its dashboard
`agentmoney usage`	read real usage from the Anthropic Usage API
`agentmoney sandbox-run`	price an untrusted call log inside an E2B microVM
`agentmoney memory-search`	BM25 search over the memory and docs of the folder you run it in

Options and settings

Flag or variable	Default	What it does
`-m, --model <id>`	required	the model, for `calc` and `estimate`
`-i, --input` · `-o, --output`	required	input and output tokens, for `calc`
`--cache-read <n>` · `--cache-write <n>`	0	cache tokens, for `calc`
`-c, --calls <n>` · `--avg-input` · `--avg-output`	required, 2000, 500	calls and tokens per call, for `estimate`
`--port <n>`	`8999`	the proxy's port
`--target <url>`	`https://api.anthropic.com`	where the proxy forwards requests
`--alert <usd>`	none	warn when a session's cost passes this
`--days <n>`	7	days to read, for `usage`
`--json`	off	JSON output (`calc`, `models`, `usage`, `sandbox-run`, `memory-search`)
`ANTHROPIC_ADMIN_KEY`	none	the Admin API key `usage` sends (an `sk-ant-admin` key in `ANTHROPIC_API_KEY` also works)
`ANTHROPIC_BASE_URL`	`https://api.anthropic.com`	where `usage` reads from
`E2B_API_KEY`	none	the key `sandbox-run` boots a microVM with

Tests and evals

npm install && npm run build
npm test          # vitest: pricing, tracker, the proxy round trip, the Usage API read, the sandbox program
npm run eval      # every surface, then 13 golden tasks against the built CLI, MCP server and hook, pass^3

No test or eval calls a real API: eval/stub-usage.mjs stands in for the Usage API. CI runs both on every push and pull request. The most useful contributions are fresh rates (follow skills/add-model-pricing), new providers such as Mistral, Cohere or DeepSeek, historical trends and budget limits.

Supported models and pricing

USD per million tokens, as of March 2026, from each provider's pricing page. Names match exactly, then by prefix (a date suffix such as claude-sonnet-4-6-20260301), then by containment (a provider prefix such as anthropic/claude-sonnet-4-6). An unknown model is estimated at $3 in and $15 out. Where a provider publishes no cache rate (a dash below), cache tokens are charged at the input rate. agentmoney models prints the same table, and the nightly loop flags the rates for a check when they are more than 90 days old.

Anthropic

Model	Input	Output	Cache read	Cache write
claude-opus-4-6	$5.00	$25.00	$0.50	$6.25
claude-sonnet-4-6	$3.00	$15.00	$0.30	$3.75
claude-opus-4-5	$5.00	$25.00	$0.50	$6.25
claude-sonnet-4-5	$3.00	$15.00	$0.30	$3.75
claude-haiku-4-5	$1.00	$5.00	$0.10	$1.25
claude-opus-4-1	$15.00	$75.00	$1.50	$18.75
claude-opus-4	$15.00	$75.00	$1.50	$18.75
claude-sonnet-4	$3.00	$15.00	$0.30	$3.75
claude-3-5-sonnet	$3.00	$15.00	$0.30	$3.75
claude-3-5-haiku	$0.80	$4.00	$0.08	$1.00
claude-3-opus	$15.00	$75.00	$1.50	$18.75
claude-3-haiku	$0.25	$1.25	$0.03	$0.30

OpenAI

Model	Input	Output	Cache read
gpt-4.1	$2.00	$8.00	$0.50
gpt-4.1-mini	$0.40	$1.60	$0.10
gpt-4.1-nano	$0.10	$0.40	$0.025
gpt-4o	$2.50	$10.00	$1.25
gpt-4o-mini	$0.15	$0.60	$0.075
gpt-4-turbo	$10.00	$30.00	-
gpt-4	$30.00	$60.00	-
o3	$2.00	$8.00	$0.50
o4-mini	$1.10	$4.40	$0.275
o1	$15.00	$60.00	-
o1-mini	$3.00	$12.00	-
o3-mini	$1.10	$4.40	-

Google

Model	Input	Output
gemini-2.0-flash	$0.10	$0.40
gemini-2.0-pro	$1.25	$5.00
gemini-1.5-pro	$1.25	$5.00
gemini-1.5-flash	$0.075	$0.30

What it is made of

Every project in this fleet is graded on the same 11 harness components and on self-improvement. Each mark is backed by a file you can open.

Component	Grade	Where
Identity	✅	identity/SOUL.md:1 (Ledger; SOUL claims fixed, BRAND, HEARTBEAT and bootstrap MEMORY rewritten, no template slots)
Memory	✅	memory/MEMORY.md:1 (about 600 tokens) + BM25 index src/memory-search.ts:40 (agentmoney memory-search) + memory/LEARNINGS.md (7 entries, each with a rule)
Skills	✅	skills/cost-investigation/SKILL.md:1, skills/add-model-pricing/SKILL.md:1, skills/isolate-untrusted-pricing/SKILL.md:1 (moved from .claude/skills, updated to the fixed code)
Tools/CLIs	✅	package.json:9 (bins agentmoney, agentmoney-mcp; prepare builds on install); src/index.ts:27 (--version); 8 commands
Hooks	✅	.claude/settings.json:4 (PreToolUse dangerous-command and secret guards) + hooks/hooks.json:3 (plugin SessionStart spend summary, hooks/spend-summary.mjs)
Sub-agents	✅	.claude/agents/cost-analyst.md:1, .claude/agents/pricing-curator.md:1 (bundled in .claude-plugin/plugin.json)
MCPs	✅	src/mcp-server.ts:44 (cost_query; answers the client's protocol version; bin agentmoney-mcp) + .mcp.json:3 (its own server through npx, nothing else)
Dispatch	✅	.github/workflows/cost-watch.yml:23 (daily: prices a session with the product, scripts/cost-watch.mjs routes on a budget gate) + .github/workflows/improve.yml:10 (nightly loop)
Evals	✅	eval/run.sh:1 (surfaces 43, golden 13 tasks pass^3, offline, Usage API stand-in) gating CI .github/workflows/ci.yml:35
Observability	✅	src/logger.ts:13 (every proxied call to .agentmoney/calls.jsonl with tokens, cost, duration) + proxy /costs and /dashboard (src/proxy.ts) + daily reports
Sandbox/Infra	✅	src/sandbox-runner.ts:63 (E2B microVM) + src/sandbox-pricer.ts:31 (the in-sandbox program, generated from the host rate table)
Self-improvement	✅	dsip.json:1 (L2) + .github/workflows/improve.yml:10 (nightly) + scripts/improve.mjs (evals, docs and skill coverage, memory, rate freshness, README drift)

✅ real and tested · 🟡 partial, with the reason in the row · ❌ absent

How it improves itself

It carries its own evals and a self-improvement contract, dsip.json, which says what it may change and how a change is judged. A scheduled job runs its evals and proposes fixes. A person makes every change. Lessons it learns are kept as rules in its memory files.

Part of a fleet

One of the projects by Naman Parikh. Each owns one part of an agent's harness, or of the product around an agent, and all of them follow one standard.

Related project	What it owns
AgentDial	Identity
WikiMem	Memory
Skillsmith	Skills
Agent Booster	Tools / CLIs
Hooksmith	Hooks
AgentCast	Sub-agents
Plugsmith	Plugins
Armory	MCPs and the component registry
AgentGrid	Dispatch
AgentSwarm	Dispatch
AgentBench	Evals
Sandforge	Sandbox / Infra
Claude Harness	The whole harness, as a template

Community

Questions and ideas go to Discussions; bugs go to issues. Follow the work on X and LinkedIn.

Contributing and security

Read CONTRIBUTING.md. AI-assisted pull requests are welcome. Report a vulnerability privately, as SECURITY.md explains.

Navigation Menu

Repository navigation

About

Repository files navigation

agentmoney

Use it from any coding agent

Quick start

Example output

How it works

The proxy

Reports

Commands

Options and settings

Tests and evals

Supported models and pricing

Anthropic

OpenAI

Google

What it is made of

How it improves itself

Part of a fleet

Community

Contributing and security

Star history

Contributors

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
.claude-plugin		.claude-plugin
.claude		.claude
.github		.github
.well-known		.well-known
assets		assets
brain		brain
docs		docs
eval		eval
hooks		hooks
identity		identity
memory		memory
scripts		scripts
skills		skills
src		src
test		test
.claudeignore		.claudeignore
.editorconfig		.editorconfig
.env.example		.env.example
.gitignore		.gitignore
.mcp.dev.json		.mcp.dev.json
.mcp.json		.mcp.json
AGENTS.md		AGENTS.md
CITATION.cff		CITATION.cff
CLAUDE.md		CLAUDE.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTEXT.md		CONTEXT.md
CONTRIBUTING.md		CONTRIBUTING.md
GEMINI.md		GEMINI.md
LICENSE		LICENSE
NOTICE		NOTICE
QUICKSTART.md		QUICKSTART.md
README.md		README.md
SECURITY.md		SECURITY.md
TODOS.md		TODOS.md
dsip.json		dsip.json
gemini-extension.json		gemini-extension.json
llms.txt		llms.txt
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
tsconfig.json		tsconfig.json
vitest.config.ts		vitest.config.ts

Latest commit

History

Folders and files

Repository files navigation

agentmoney

Use it from any coding agent

Quick start

Example output

How it works

The proxy

Reports

Commands

Options and settings

Tests and evals

Supported models and pricing

Anthropic

OpenAI

Google

What it is made of

How it improves itself

Part of a fleet

Community

Contributing and security

Star history

Contributors

License

About

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages