AI-d kit

About this pool

Tools for choosing, pricing and debugging AI models, built while running a fleet of agents. All of it is free and none of it needs an account.

Open Dashboard MCP

Ask your agent which model to use and what it will cost, and get an answer built from live provider catalogues. It is read-only. A missing key or a blocked provider comes back as exactly that, never as an empty or free result.

Add it to Claude Code:

claude mcp add open-dashboard -- npx -y open-dashboard-mcp

Latest version on npm (external site)Browse the dashboard

19: read-only tools, from price comparison to what changed this week
16: providers, each catalogue read live from the provider's own source on every call: OpenRouter, Groq, Cerebras, Sail, Nous Research, QwenCloud, DeepInfra, Novita, SambaNova, Chutes, WaveSpeedAI, fal, KIE, Crazyrouter, AkashML and io.net. Twelve need no key; AkashML, Groq, Sail and QwenCloud need their own.
Per model: prices with their source and read time, and a missing price reported as unknown, never zero. Price reads were measured on 26 September 2026 for 7 of the 16 providers: AkashML 6 of 6 models priced, io.net 37 of 37, Groq 8 of 11, OpenRouter 452 of 458, Sail 12 of 12, Cerebras 2 of 3, Nous Research 417 of 417, marked as a reseller of OpenRouter's catalogue rather than an independent source.

Two references, both free

Troubleshooting

108 failures that reported success, or reported absence when something was there. Every entry is written the same way:

Symptom: what you saw
Cause: what was actually wrong
Cost: the hours it took
Check: the probe that would have caught it

Free stuff in promotions

Genuinely free AI credits and standing free tiers. Each offer shows the day it was last checked against the provider's own page, and nothing unverified is shown as live.

Search first, then read

The unified search and glossary are the main way in. Mapping a blocker to a tool is a guided mode inside the pool, not a gate in front of it.

LibraryAI terms, platforms, tools and research collections in one searchable index, glossary included.
Explore ReposEvery public repository on the fleet's GitHub account, originals and reference forks, searchable in one place.
Repo Shelf (external site)1,268 public repositories in the Explore Repos snapshot, generated 2026-10-04 from the verified 2026-10-02 cutoff, as books on a 3D shelf to browse rather than search.
The Drop (external site)A daily AI publication. Today's pick first, then what still matters from earlier drops.
Model councilsPut one question to a panel of free models. Public and stateless: nothing you ask is remembered.

In development

Built and under review, not public yet. These get links when they ship.

Model Router: An OpenAI-compatible proxy that picks the model for each request with a typed, readable decision, or refuses.
Glass PR Solver: Drains a pull-request backlog through four independent stages, with every fact and decision visible.
Glass Council: A private model council you install yourself, on Windows, Android or as an MCP server.
System One Bench: Measures per-decision latency and cost of typed decision models on your own machine, and says UNAVAILABLE rather than comparing when a provider is missing.

The AI-d kit catalog

Every project the site catalog lists for this pool, each with its status and review record under "Catalog record". The in-development projects above are not catalogued yet.

Project details load from the shared catalog. If they do not appear, you can still explore the other pools below.