Cut token spend 30–60%.Keep the quality.
Optimize runs as a lightweight sidecar next to each developer's coding agent, steering model, effort, and context decisions in the background. Same engineers, same tools, same work. A 5-minute install, and a paired A/B test that proves the savings on your own repository.
No behavior change for developers
One-time install
brew install codelexica/tap/cdlx, then cdlx init. About 5 minutes per developer, once.
Same tools, same workflow
Claude Code and Codex keep working exactly as before. No new IDE, no new agent, nothing to learn.
Savings you can verify
Every number is measured against your own team's usage, and a paired A/B test proves it on your tasks before you commit.
Cut token spend 30–60%. Keep the quality.
Optimize runs as a lightweight sidecar next to each developer's coding agent. It steers model, effort, and context decisions in the background, so the same engineers ship the same work on far fewer tokens.
Optimize (Standard)
Without access to your codebase
up to 30%
typical token-spend reduction
Meaningful savings from day one, with nothing indexed and no code leaving your machines.
Model routing. Each session runs on the cheapest model that can actually do the job, decided from task difficulty, not habit.
Effort steering. Reasoning effort is matched to the task, so routine work stops paying for deep-thinking budgets.
Subagent model pinning. Helper agents run on efficient models while the lead agent keeps the frontier model.
Session budgets. Per-session token budgets with org-level defaults, so a runaway loop cannot burn a day's spend.
Context management. Constraint narrowing, structured handoffs, and cross-provider continuity keep sessions lean without dropping what the agent needs.
Optimize (Premium)
With your codebase indexed on Code Lexica
30–60%
typical token-spend reduction
Everything on the left, plus persistent knowledge of your architecture and conventions, so agents stop paying to rediscover it on every run.
Advanced routing. Cheaper models succeed far more often when they are handed the right context, so more work moves down-market safely.
Just-in-time context. The architecture, conventions, and exact files that matter are injected automatically at session start and per task. No prompting, no copy-paste.
Agent-ready specifications. Task specs grounded in your codebase, written for the agent, so it starts implementing instead of exploring.
Persistent codebase knowledge. A living model of your repositories that every agent session draws on, kept current as the code changes.
Two ways to prove the savings
One-off · Controlled
Paired A/B test
A controlled experiment on your own repository, finished in an afternoon.
cdlx ab-test run --task "…"Paired worktrees. Same task, same base commit, in isolated worktrees: one arm with Code Lexica, one without.
Fair pricing. Both arms priced from one rate table, so API-key and subscription usage compare fairly.
Full scorecard. Tokens, cost, duration, and verify-command pass rate per arm.
Day-one results. Measured savings land on your Savings dashboard before any baseline history exists.
Ongoing · Real-world
Ongoing Savings Tracker
Sample DataLive savings on your team's real work, measured against your own pre-Code-Lexica baseline.
Est. savings (API-rate value)
$3.2K–$4.1K
vs your team's usage before Code Lexica
Tokens / message
−31%
1,250 with · 1,800 without
Code Lexica adoption
9/12
devs using Code Lexica this window
Monthly spend, same team
$16.2K before $13.8K with Code Lexica, at higher volume
5-Minute One-Time Setup
Install
Install the Code Lexica CLI tool via Homebrew.
brew install codelexica/tap/cdlxTrack
Connect to your Code Lexica account and select repos to track.
cdlx initOptimize
Enable token optimization to run in the background.
cdlx sidecar enableMeasure
Measure real savings live on actual coding tasks in your codebase.
cdlx ab-test runFrom pilot to steady state
Pilot
DaysOne team installs the CLI and runs Track. A paired A/B test on your own repository gives you a measured number before anything is rolled out.
Rollout
WeeksOptimize is offered to every developer as the last step of init. As coverage grows, the dashboard compares each developer against their own pre-Code-Lexica usage.
Steady state
OngoingOrg policy sets routing, effort, and budget defaults centrally. Savings are reviewed monthly against a baseline that keeps learning from your usage.
What leaves your machines
Sent to Code Lexica
- Token counts and model identifiers
- Tool-call names and result status
- Session timings and repository fingerprints
- Optimize intervention counters
Never stored
- Prompt or response text
- Tool arguments
- File contents
- Project paths (hashed by default)
Indexing your codebase for the second Optimize tier is a separate, explicit step. A/B-test task descriptions are stored because you write them for the test.
SOC 2 Type 1 certified; security program aligned to NIST 800-53 Rev 5. SSO/SAML, role-based access, encryption at rest. Read the full security overview.
Questions, answered
If you don't save, you don't pay
Optimize is priced against the savings it creates for your organization. If we can't save you money, we'll refund you. Start a free 21-day trial today to see the savings yourself.