Ask a coding agent to pick a database and it likely chooses from training data that’s six to twelve months stale, if not older. At best, it uses a less intelligent agent to search the wilds of the internet and comes across marketing pages written for humans and Stack Overflow answers from 2022. It will confidently recommend a deprecated library. It will hallucinate package names. It will pick whatever it sees most often in training, which is how popularity became the world’s worst tool-selection algorithm.
The Zaira Guide is what we built instead: structured developer tool intelligence for over 900 tools across 60 categories and growing, written for the thing actually making the decision.
What an entry shows
Every tool page answers the questions an agent asks mid-workflow, in a form it can parse:
- Decision logic. When to use this tool, when to avoid it, and what it’s best for. Specific conditions, not adjectives.
- Alternatives and integrations. What to consider instead and why, what it works with, what it conflicts with. Over 6,000 verified cross-references across alternatives, integrations, and conflicts, every one resolving to another Guide entry.
- Constraints an agent can filter on. Pricing model, real free tier or not, license, deployment models, self-hostability, edge compatibility, SDK languages, MCP support, compliance certifications.
- Live health signals. Stars, downloads, release recency, maintainer concentration, OpenSSF Scorecard, security advisories. Pulled from the source APIs, not remembered from training.
- A Zaira Base Score with its breakdown. Every tool is scanned against the Base Standard’s 15 criteria, and the score splits into three sections (documentation and usability, provenance and disclosure, lifecycle health) so a 63 score tells you where the 63 comes from.
- Implementation FAQs. The questions an agent asks after choosing, answered from evidence.
One query, for real
Say an agent gets asked for a relational database with a real free tier for per-PR preview environments. Here’s the actual call, live right now:
GET https://zairalabs.ai/guide/api/v1/tools?category=relational-database&hasFreeTier=true&limit=100
→ 12 results. Each is a shortlist card: name, tagline, category,
pricing model, free-tier flag, Zaira Base Score.
GET https://zairalabs.ai/guide/api/v1/tools/{slug}
→ one more call per candidate returns the full decision block:
use_when · avoid_when · best_for
alternatives (with reasons) · works_with · conflicts_with
min_free_tier terms · license · self_hostable · edge_compatible
mcp_support · ard_catalog · compliance
zaira_score + section breakdown · last_verifiedFrom there the agent isn’t reading marketing pages. It’s filtering decision logic: which candidates name branch-based preview environments in their use_when, which get ruled out by an avoid_when constraint, what the alternatives blocks recommend for this workload. Shortlist, pull two or three full entries, decide. Over MCP it’s tighter still: zaira_search_tools folds the decision logic straight into the search results, and zaira_compare_tools returns full entries side by side in one call.
Run the query yourself. The count, the fields, and every score in the response are live production data, and the same decision block is on every one of the 900+ tool pages.
Say nothing rather than guess
The Guide’s core rule is simple: a fact we can’t verify is a fact we don’t publish.
Every entry is assembled by a pipeline that collects from dozens of public sources, resolves each field through a trust-ordered ledger, and then verifies the facts a model might otherwise guess against the collected evidence. A license comes from the repository’s LICENSE file, not from prose. A free tier gets published only when the pricing page confirms a standing allowance a new customer can get today. An MCP server link ships only after we’ve fetched it and confirmed it’s a live, first-party server, because pointing an agent at a dead endpoint is worse than saying nothing.
When verification fails, the field is omitted. Internally, every published claim traces back to a per-field decision record with its source and evidence. That’s also why entries carry a last-verified date instead of pretending to be timeless.
What the Guide doesn’t know
The honesty rule cuts both ways, so here’s what you won’t get. Closed-source managed services publish no repository, so their entries carry no stars, contributor counts, or OpenSSF Scorecard: those fields sit empty rather than estimated. Pure open-source tools carry no free-tier flag at all, because a free tier means nothing when the whole product is free; in the Guide, null means not applicable, never a guess. Base Zaira scores measure fundamentals from public evidence, so a tool with undocumented strengths scores as if it lacked them, and a score can shift a few points between scans as a vendor’s public surfaces change. A missing field can drag a score, but not always: scoring is lane-aware between open-source and closed-source tools and adjusts what evidence it expects. As things change, and as tools get better data, the Guide is updated on every rescan.
One dataset, three channels
Agents don’t discover things one way, so the Guide doesn’t serve one way.
Web. Nearly 4,000 pages structured for machine reading: answer-first sections, tables for structured data, JSON-LD on every page. This is how answer engines find us, and they already do. During the quiet soft launch, ChatGPT user-generated queries alone referenced Guide pages hundreds of times a day.
REST API. GET /guide/api/v1/tools?category=relational-database&hasFreeTier=true&selfHostable=true filters on any constraint and returns structured JSON. Request any single entry with Accept: text/markdown and the same data comes back as Markdown-KV, the format LLMs parse most reliably. No key, no auth.
MCP. Install the Zaira MCP server (https://zairalabs.ai/guide/mcp) and your agent queries the Guide as a tool: zaira_search_tools by constraints, zaira_get_tool for a full entry, zaira_compare_tools for side-by-side candidates, all mid-session. A few minutes to add it.
Same data, same verification, whatever surface your agent already lives on.
Go use it
The Guide is free. No placement is for sale, no vendor money changes a score, and the methodology is public. If the data in your entry is wrong, tell us. The fastest durable fix is usually ensuring your public docs are accurate and queryable, and the Zaira Standard defines exactly what that looks like for agents.
If your tool isn’t in there, drop it in the queue. We are constantly adding tools to the guide and if yours fits the bill you will probably see it in there.
Let’s make building better things easier. Start by giving agents the data and trust signals they need so the humans get the results they want.
zaira_labs