Optimize how your
agents run.

LeanTokens desktop · interactive preview click to explore

LeanTokens routes each task to the right model, harness, tools, and context.

Request early access → See how it works ↓
Agent stacks are getting more complicated. LeanTokens is the command center that optimizes token usage and cost.

Every run is a set of decisions.

LeanTokens breaks down a task into subtasks and selects an execution setup for each: agent, model, reasoning effort, tools, skills, and context. It then verifies the result and records the decision.

LeanTokens control flow A prompt from LeanTokens and any connected framework is planned into four branching sub-tasks. t2 itself branches further, to the right, into t2a and t2b. Each sub-task is configured with a harness, model, and context, then merged into a verifier before output. Output feeds a policy-update loop back to the planner. Prompt LeanTokens Planner decompose · assign t1 · Luna 2k context t2 · Sonnet 3k context t3 · Sonnet 8k context · qa.skill t4 · Opus 4k context t2a · Luna 1k context t2b · Opus 2k context Verifier quality gate Output tokens · latency · cost LEARNING LOOP · POLICY UPDATE ITERATE
one prompt in, branching sub-tasks configured, one verified output
policy updates after each run ↑

Built around the tools you already use.

Harnesses
Models
GPT-5 · Claude Sonnet · Claude Opus · Gemini · Approved internal models
Controls
Per-run cost caps, latency targets, minimum accuracy, maximum retries.
reports
Every run is reproducible, diffable, and explains why it chose what it chose.

Optimize for cost, speed,
and quality.

01 ↓
COST
Only the tokens you need.

Route each sub-task to the smallest capable model. Size context to the step. Skip redundant passes. Cache and reuse where valid.

02 ↓
LATENCY
Parallel by default.

Independent steps run in parallel. Long chains collapse where they can. Verifier retries are bounded, not open-ended.

03 ✓
QUALITY GATE
A floor you set.

You set an accuracy floor. Cost and latency optimizations only ship if outputs still pass the verifier and your regression checks.

Keep your agent stack. Add a control layer.

fig 02 . positioning
concern
model routers
Harnesses
leantokens
Model routing per sub-task
✓
✓
✓
Prompt + context sizing
✓
✓
Agent execution + tools
✓
✓
Verifier-guarded outputs
✓
✓
Joint quality + tokens + latency
✓
Every run makes the next cheaper and better
✓

Plan, configure, verify, learn.

fig 03 . one run
leantokens . orchestrate.log ● LIVE
$ leantokens run "extract + summarize"
 
plan │ 3 sub-tasks . parallel: 2
config│ t1 extract Luna . context 2k
│ t2 classify Luna . context 1k
│ t3 summarize Sonnet . context 8k
verify│ t1 ok t2 ok t3 retry → ok
learn │ Luna @ extract, saved 62% tokens
────────────────────────
done │ q 0.94 tokens 11.2k t 3.8s