OUTPACE
Get more from your token limits. And your budget.
Your team hits caps before the work is done. Palindrome finds the same work in fewer tokens: cheaper-equivalent models, leaner prompts, tools you pay for but don't use. Palindrome recommends, you decide.
Palindrome — Recommendationsacme-eng
Inbox12 open
Sorted by $ impact1
Switch Q&A prompts to Sonnet 4.6TOP PICK
model swap96% matcheffort: low
−$340per month
2
Drop filesystem MCP from default toolset
unused tool91% confeffort: low
−$120per month
3
Compress qa-system prompt (−67% tokens)
prompt97% matcheffort: med
−$210per month
4
Cache shared retrieval preamble
caching93% confeffort: low
−$95per month
8 more in queueIf applied:−$765/mo · −41% tokens
Use a cheaper-equivalent model
Pattern-level equivalence with confidence scores. The same output, fewer premium-model tokens.
Advice stretches limits by lightening the work. It doesn't raise your vendor's caps. You get more from your limits.
Q&A prompts · model swap−$340/mo
CURRENTClaude Opus 4.8$15.00 / 1M input tokp50 11.2s · 1,840 tok/req
PROPOSEDClaude Sonnet 4.6$3.00 / 1M input tokp50 4.8s · 1,840 tok/req
96% structural match · eval Δ −0.4% · within toleranceView eval run →
Idle context · last 14 days
$265/mo reclaimablefilesystem MCPunused 14d
2.1k tok/req
−$120/mosql-runner MCPduplicate of db-query
1.4k tok/req
−$85/molegacy-rag toolsuperseded 31d ago
0.9k tok/req
−$60/moTotal−$265/mo · −4.4k tokens per request
ReclaimTrim what you don't use
Unused or duplicative MCP servers and tools quietly burning context on every single request.
Smarter prompts
Caching opportunities and shared canonical patterns that shrink token footprints across the team.
qa-system-prompt.md
1,842→612 tokens−67%
@@ -1,42 +1,11 @@
− You are a helpful, friendly QA assistant. Always answer politely and clearly.
− Before answering, restate the question and summarize the relevant policies.
− Return JSON with keys answer, tone, confidence, citations, follow_up, notes.
+ include: @canon/support-voice (cached)
+ include: @canon/output-schema
+ context: policy_chunks(top_k=3)
97% output match·cache hit 84%
−$210/moSavings report — May
−$2,140this month
baseline forecast $9,860 → actual $7,720−21.7% vs baseline forecast
Weekly spend
BaselineActual
W1
W2
W3
W4
synthetic-reconstruction validation · 96% structural match
Export evidence (CSV)Savings you can defend
Every confidence number is backed by a synthetic-reconstruction validation methodology, not vibes.
Ranked by $ impactConfidence + effort on every recAdvisory. Never silent changesValidated equivalence methodology
Find your savings this week.
Free for design partners · We value your privacy