Track what your AI agents actually cost per task, not per month

Brian Munene Mwirigi — Mon, 09 Feb 2026 13:15:12 +0000

## The problem

Your AI agent just finished a 45-minute coding session.  
It edited 15 files, made 4 commits, and called the API 23 times.

What did it cost?

No idea. You’ll find out on your next invoice aggregated with everything else.

## The fix

I built **codesession-cli** — a CLI that teaches your AI agent to track its own costs.

## Setup (30 seconds)

```

bash
npm i -g codesession-cli
clawhub install codesession

Start a new OpenClaw session. Done.

What happens

The agent reads the skill instructions and handles everything.


bash
cs start "Fix auth bug" --close-stale --json
# → agent works on your task normally…

cs log-ai -p anthropic -m claude-sonnet-4 --prompt-tokens 8000 --completion-tokens 2000
# → logs its own token usage, cost auto-calculated

cs end -n "Fixed bug, added tests" --json
# → Session: 9m • 3 files • 1 commit • $0.15

You don’t touch anything. Just review the data when you’re curious.


bash
cs stats
# Total: 50 sessions • 8h 34m • $47.23 AI cost

What’s tracked

Token spend per API call (17+ models with built-in pricing)
File changes and git commits (via git diff, not a watcher)
Session duration and cost summary
Annotations — the agent leaves breadcrumb notes as it works

Tech

TypeScript
SQLite (WAL mode)
Local storage (~/.codesession)
JSON output on every command — parse schemaVersion for forward compatibility
Structured errors: { error: { code, message } }, always exit 1
Sessions scoped by git root, not cwd
MIT licensed

have you guys tried it?

Brian Munene Mwirigi — Sun, 01 Feb 2026 19:03:31 +0000

I Got Tired of Burning Money on AI APIS - Built a CLI Tool to Fix It - Vibe Coding Forem

I checked my OpenAI billing last month. $217. I had no idea where it went. The solution: I built...

vibe.forem.com

I Got Tired of Burning Money on AI APIS - Built a CLI Tool to Fix It

Brian Munene Mwirigi — Sun, 01 Feb 2026 19:02:10 +0000

I checked my OpenAI billing last month. $217. I had no idea where it went.

The solution: I built aitoken-cli. Now I track every cent.

The result: Reduced spending to $89/month. 59% savings.

The Hidden Cost of AI Development

If you're a developer in 2026, you're using AI APIs. GPT-4, Claude, Gemini they're essential tools.

The costs creep up silently.

My Wake-Up Call

December 2025:

Week 1: $12 (seems fine)
Week 2: $28 (okay, working on a project)
Week 3: $51 (wait, what?)
Week 4: $126 (WTF?!)

Total: $217

I was a broke CS student burning $217/month on AI. Something had to change.

Why Existing Solutions Failed Me

You'd think tracking API costs is solved, right? Wrong.

OpenAI Dashboard

Shows totals only
No breakdown by task
No comparison across providers
Have to log in every time

Spreadsheets

Manual entry (I forgot constantly)
No automatic cost calculation
Boring, never kept it updated

Existing npm packages

Too complex (need servers, databases)
Privacy concerns (send data to cloud)
Not comprehensive (OpenAI only)

The gap: I needed something local, fast, and comprehensive.

Building aitoken-cli

Friday night, 9 PM: Just got my $217 bill. Angry.

Friday night, 11 PM: Sketched the design.

Saturday, 3 AM: First version working. Tracking OpenAI only.

Saturday, 9 AM: Added Anthropic, Google, Azure, Cohere.

Saturday, 2 PM: Published to npm.

Total time: 6 hours.

How It Works

Track Usage in Seconds

# Track a GPT-4 call
at add -p openai -m gpt-4o -i 1500 -o 800

# Output: ✓ Added usage #47 - openai/gpt-4o - $0.0118

View Your Stats

at stats

# Output:
# Total Usage: $34.56 (245,890 tokens)
# 
# By Provider:
# OpenAI:    $21.34 (150,000 tokens)
# Anthropic: $11.22 (85,000 tokens)
# Google:    $2.00  (10,890 tokens)

Today's Spending

at today

# Output:
# Today's Usage: $4.23 (15,670 tokens)
# 
# Latest entries:
# 10:15 AM - gpt-4o     - $1.52 (5,000 tokens)
# 11:30 AM - claude-3.5 - $2.71 (10,670 tokens)

Integrate with Your Code

// After an OpenAI call
const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: messages
});

// Track it
const { prompt_tokens, completion_tokens } = response.usage;
exec(`at add -p openai -m gpt-4o -i ${prompt_tokens} -o ${completion_tokens}`);

The Impact: Real Savings

Before aitoken-cli (December 2025):

Monthly cost: $217
Awareness: Zero
Optimization: None

After aitoken-cli (January 2026):

Monthly cost: $89
Awareness: 100%
Optimization: Strategic model switching

Savings: $128/month (59% reduction)

How I Optimized My AI Spending

Discovery 1: GPT-4 for Everything Was Wasteful

Before: Used GPT-4 for simple tasks like string formatting

Fix: Use GPT-3.5 Turbo for simple tasks

Savings: $40/month

Discovery 2: Long Prompts Every Time

Before: Sent full system prompts with every request

Fix: Cache system prompts, send only deltas

Savings: $30/month

Discovery 3: Claude is Often Better AND Cheaper

Before: Default to GPT-4 for everything

Fix: Use Claude 3.5 Sonnet for coding tasks

Savings: $35/month

Discovery 4: Redundant API Calls

Before: Making duplicate calls without noticing

Fix: Added request deduplication

Savings: $23/month

None of this was possible without tracking.

Technical Details

Tech Stack

TypeScript - Type safety for pricing data
Commander.js - CLI framework
SQLite (better-sqlite3) - Local database
Chalk - Pretty terminal output

Database Schema

CREATE TABLE usage (
  id INTEGER PRIMARY KEY,
  provider TEXT NOT NULL,
  model TEXT NOT NULL,
  prompt_tokens INTEGER NOT NULL,
  completion_tokens INTEGER NOT NULL,
  total_tokens INTEGER NOT NULL,
  cost REAL NOT NULL,
  timestamp TEXT NOT NULL,
  notes TEXT
);

Simple. Effective. Local.

Privacy: Why Local Matters

aitoken-cli stores everything in ~/.aitoken/usage.db.

No cloud. No API. No tracking.

Why?

Your usage patterns are private - Nobody knows what you're building
Works offline - No internet? Still tracks
Fast - No network latency
Free forever - No subscription costs
You own your data - Export anytime

Installation

npm install -g aitoken-cli

Quick Start

# Track your first API call
at add -p openai -m gpt-4o -i 1000 -o 500

# View your stats
at stats

# See today's usage
at today

# List all providers and models
at models

What's Next

Coming soon:

Budget alerts (warn when you hit $X/month)
Model recommendations (suggest cheaper alternatives)
CSV export for accounting
Team usage tracking

Try It Today

Challenge

Track your AI usage for one week. I bet you'll be surprised by what you find.

npm install -g aitoken-cli

Then tweet your results with #aitokencli – I'll retweet the most interesting findings.

Resources

Other Tools I've Built

runbook-cli - Remember project commands
codesession-cli - Track coding sessions

Building in public: Follow my journey @brian_mwirigi
Github: github.com/brian-mwirigi

Blog: brianmunene.me/blog

INFO: If you find any bugs or issues please feel free to reach out
`

Forem: Brian Munene Mwirigi