TCalc

Open-source. Local-first.
Built for your workspace.

Star on GitHub

Know your context.
Choose the right model.

TCalc analyzes your code and prompts locally to calculate accurate token counts and recommend the best coding models for the job.

Install for VS Code Read the docs
  • 100% local
  • No telemetry
  • Open source
Workspacetcalc / srcModel profileCoding (General)ScanJust now
TOKEN SUMMARY1,246,512
862,132Input tokens384,380Output tokens
CONTEXT BREAKDOWNCode72.4%Comments12.8%Docs8.7%Other6.1%
Files 236Lines 18,392Languages TypeScript, Python
RECOMMENDED MODELS
  1. 1
    Gemini 2.5 ProBest fit1,048,576 context
    $0.86est. cost
  2. 2
    GPT-4o128,000 context
    $1.32est. cost
  3. 3
    Claude Sonnet 4200,000 context
    $1.18est. cost
  4. 4
    Mistral Large 2128,000 context
    $0.74est. cost
  5. 5
    DeepSeek V3128,000 context
    $0.52est. cost
Pricing cached: 2m agoConfigure models →
Supports leading providers
+ more

Features

  • Accurate token countingModel-aware tokenization for precise estimates.
  • Smart model recommendationsRanked by fit, context window, and cost.
  • 100% local and privateEverything runs on your machine. Zero telemetry.
  • Workspace awareRespects .gitignore, files, and language heuristics.

Workflow

  1. 1
    Open a workspaceTCalc scans your project locally.
  2. 2
    We count your contextInput, output, and file breakdowns.
  3. 3
    Get model recommendationsSee the best models for your context and budget.
  4. 4
    Pick and shipChoose with confidence and keep building.

Documentation

View all docs →

Local report viewer

Open the evidence.
Keep it in your browser.

Select a JSON report generated by TCalc. It is parsed on this page and is never uploaded to the hosted site.

Showing a safe example report.

add-feature · 7/20/2026

commerce-platform

Included tokens184,760Files scanned742Files excluded130
  1. 01
    DeepSeek V3Cheapest sufficient · DeepSeek
    86/100
  2. 02
    Gemini 2.5 ProBalanced · Google AI
    91/100
  3. 03
    Claude Sonnet 4High confidence · Anthropic
    89/100

Common questions

Clear answers,
before the scan.

Does TCalc upload my source code?

No. Workspace scanning, token estimation, model ranking, repo maps, and report generation run locally. The hosted report viewer reads a selected JSON file in your browser.

Why might only one model fit?

The required context, local-only privacy mode, team model policy, or a small custom catalogue can narrow the eligible candidates. The model-catalog guide shows how to diagnose each filter.

Can I use TCalc outside VS Code?

Yes. The monorepo includes a CLI, GitHub Action workflows, an optional report service, and an experimental stdio MCP server for coding agents.