[ 05 — 3 live calculators · 5 downloadable worksheets ]
Less guessing. More building.
Small instruments for better decisions. Estimate a bill, inspect labeled results, plan memory or write an agent's operating contract. These tools do real, bounded work—without calling a model or inventing a benchmark.
Token Thermostat
A cost workbench with explicit cache buckets and pricing conditions.
Estimate a text-inference workload using the source-checked model rates. Adjust ordinary input, cache reads, cache writes, output and request count. See the total and the assumptions, including GPT-6 long-context multipliers, Gemini introductory pricing and DeepSeek peak/off-peak conditions.
Server-side
Calculation
None
LLM calls
Not saved
Input storage
Eval Result Analyzer
Summarize your labeled runs without inventing a model winner.
Paste already-scored JSON records and calculate attempts, accepted outcomes, average latency, total cost and cost per accepted result for each candidate. Failed attempts remain in the bill. The sample uses fictional Candidate A and Candidate B labels and is explicitly illustrative.
500
Max records
Labeled runs
Input
Not saved
Retention
Local Memory Planner
Understand the lower bound before downloading a large checkpoint.
Estimate raw weight storage from parameter count and precision, add an explicit planning allowance and compare it with available memory. See both decimal GB and binary GiB. The tool deliberately avoids promising that a model fits or will run quickly.
2–32 bits
Precision
Lower bound
Estimation
None
Telemetry
Agent Brief Builder
Give an agent a task, a boundary and a stop condition.
A reusable execution contract for a bounded agent task. Fill in the goal, input sources, allowed tools, approval gates and acceptance checks. Download it for your repository or workflow tool; it is not an autonomous runtime and never executes commands.
Markdown
Output
None
Execution
Yes
Editable
Evidence Ledger
Keep claims, citations, dates and caveats in separate columns.
A CSV template for researching current AI releases or tracking citations in answer engines. Use one row per claim or observed citation. It distinguishes a mention from a supporting source and keeps the publication date separate from your review date.
CSV
Output
Labeled
Example
Manual
Tracking
Voice → Product Brief
Preserve decisions and open questions when converting notes into a spec.
An editable brief template for a reviewed transcript or meeting note. The fields separate established facts, proposed scope, non-goals and unresolved questions. Use it with an approved transcription or writing tool; this page does not upload or process audio.
Markdown
Output
None
Audio upload
Human-reviewed
Decisions
Context Policy Kit
Define authority, freshness, permissions and budgets before retrieval tuning.
A JSON planning document for an evidence-backed assistant. It makes source priority, expiry rules, context allocation and abstention explicit. The schema is provider-neutral and designed for human review, not a drop-in runtime configuration.
JSON
Output
Neutral
Provider
Review first
Deployment
Release Review Sheet
Make a model or application upgrade a documented decision.
A deployment review worksheet that connects candidate configuration, evidence, permissions, cost and rollback. Use it for a model change or an AI-generated app launch; it does not generate a winner or claim a completed test.
Markdown
Output
Human-owned
Decision
Required
Evidence
Calculators run deterministic arithmetic on this application's server and do not store their input. Worksheet edits stay in your browser tab. No external AI account, API key or paid inference is required.