tokn-ai

Open source / local first

Build, route, and measure AI systems.

Tokn AI makes focused tools for people working close to language models—from training experiments to local inference and usage analysis.

Explore projectsView source
tokn-ai / terminal
# install the usage analyzer
$ npm i -g @tokn-ai/llm-tokei

$ llm-tokei --month -h

SOURCE       TOKENS      COST
codex        28.4M      $31.20
claude       12.7M      $18.44
copilot       6.1M       $7.03
────────────────────────────
total        47.2M      $56.67

Projects

Three parts of the workflow.

Independent, composable tools. Use one, or connect them into your own model-building loop.

01research preview

flm.py

A composable Python workspace for language-model research, training, reinforcement learning, and inference.

  • PyTorch foundations
  • Reusable model modules
  • Portable training experiments
PythonRepository
02in development

tokn-router

One local, OpenAI-compatible endpoint for supported coding plans and model providers.

  • Single Rust binary
  • Streaming + account pools
  • Local SQLite usage records
RustRepository
03available

llm-tokei

Fast, local reports that show where your coding-agent tokens and estimated costs went.

  • Six agent sources
  • Human tables + stable JSON
  • No log uploads
RustRepository

Why Tokn AI

Small tools, understood end to end.

We prefer software you can inspect, run locally, and compose yourself over platforms that need to own the entire workflow.

01_local_first

Keep sensitive state close.

Logs, credentials, usage records, and experiments stay on your machine by default.

02_open_source

Inspect the implementation.

Understand behavior and tradeoffs directly from the code—not through a platform promise.

03_purpose_built

Do one job clearly.

Focused commands and libraries that fit into existing systems without requiring a migration.

Available now

Start by measuring your agents.

$ npm i -g @tokn-ai/llm-tokei