Agent token optimiser

Use
less
tokens.

AI agents burn context windows reading files in full. Tokenless compresses files before the agent sees them and adds durable project context. Runs locally, works with any agent, no subscription.

Buy Tokenless See use cases
Savings by file type Running
Loaded
0%
Saved
0%
0
tokens saved
0%
savings rate
CODE
DOCS
DATA
LOGS
WEB
CONFIG
Works with your agent
Claude Claude
Codex Codex
Cursor Cursor
Windsurf Windsurf
Copilot Copilot
OpenClaw OpenClaw
Hermes Hermes
Why Tokenless

Fast. Local.
Yours.

Tokenless sits locally between your agent and your files — no cloud, no model calls, no lock-in.

Performance
Rustnative
Built in Rust

A compiled native binary, not a heavyweight runtime. Compression is deterministic and near-instant, so it never becomes the slow step in your agent loop.

Privacy
0APIs
Runs on your machine

Everything happens locally. No external APIs, no telemetry, no model calls — your code and documents never leave your device, and it works fully offline.

Pricing
1×one-time
No subscription

Tokenless is a single one-time purchase. Buy it once, then install the full local build with code, documents, data, logs, web, and memory included — no recurring bill.

Compatibility
MCPstandard
Works with any agent

Exposed over the Model Context Protocol, so Claude, Cursor, Windsurf, Copilot, Codex and more pick it up with no code changes. Cross-platform on Windows, macOS, and Linux.

Coverage
20+file types
Every file & workflow

One install covers code in any language plus PDF, DOCX, XLSX, PPTX, JSON, CSV, YAML, logs, and web pages with schema views, log filtering, and dedup.

Knowledge
Memorylocal
Unified project context

Source-backed pages and cross-session memory let agents load, search, and reuse compact project context instead of re-reading raw sources every session.

Use Cases

Four tasks.
Same tool.

Representative agent workflows covering code, data, ops, and knowledge work. Savings measured against the realistic raw-read baseline — full table in benchmarks below.

Developer

Debug a production error

Find a bug in a codebase the agent hasn't seen. Without compression, every related file is read in full.

1Search files related to the error, read at signature level
2Grep for the stack trace, expand only the failing function
3Edit without re-reading the file
77% fewer tokens vs. reading all related files in full
Data

Extract from a JSON API export

A 100-record paginated response would consume most of the context window on a single call.

1Compress to a schema view — fields and types only
2Identify the fields relevant to the task
3Pick only those fields from the full file
98% fewer tokens vs. loading the raw JSON file
Operations

Diagnose a production incident

The application log from the last hour is 50,000 lines. Loading it raw isn't feasible in any context window.

1Filter log to ERROR / WARN, dedup repeated lines
2Grep matches for the specific service or component
3Build a timeline from the surfaced events
89% fewer tokens filtered to ERROR/WARN, deduped
Documents

Query a long PDF, DOCX, or XLSX

A 60-page report or multi-sheet workbook. Reading the whole file to answer one question burns the context window.

1Extract structural text — pages, sheets, slides as headings
2Index into addressable sections with token costs
3Query the cached parse, drill in only if needed
85% fewer tokens single targeted query vs. full read
Benchmarks

Real numbers.
Real files.

Measured against realistic fixtures with the cl100k_base BPE tokenizer (the encoding behind GPT-4). Numbers below are real — regenerated by cargo test --test landing_benchmarks. No synthetic gains, no chars/4 hand-waving. The purchased install covers code, documents, data, logs, web, and config workflows.

Task Full Tokens Savings
0%
average token savings across all benchmark fixtures
20+
languages and file types auto-detected — no configuration
100%
deterministic — no model calls, works offline
Buy and install

Buy once.
Then install.

  • Pay first on the landing page, then return to a platform-specific install view.
  • Windows gets a PowerShell install command plus a direct .exe fallback.
  • macOS and Linux get direct binaries and one-time setup instructions.
  • The downloaded build already includes the full Tokenless feature set.
Continue To Checkout
Checkout returns you to an install page with Windows, macOS, and Linux downloads for the purchased build.