Axiom is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Axiom API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.
SCORED UNDER RUBRIC 1.2 · A full re-launch under Discry Score 2.5 — a new behavioral instrument, not comparable to these scores — is in progress.
Discovery
45% of score · 64/100Comprehension
55% of score · 84/100What we found
- An agent can consume Axiom's docs natively: every page is available as clean markdown via a .md suffix AND Accept: text/markdown content negotiation, plus a 94KB /docs/llms.txt index and a 3.6MB /docs/llms-full.txt full dump — this is best-in-class documentation delivery for machines.
- Axiom explicitly positions itself as 'agent-native' and ships an official hosted MCP server (axiomhq/mcp), listed on Glama and the Databricks Marketplace, so an agent can query logs/traces/metrics via APL without touching the REST API at all.
- Discoverability is split: the agent-ready assets live under /docs (llms.txt, llms-full.txt, docs sitemap), while the root axiom.co has no llms.txt and no /docs entries in its sitemap — an agent probing the apex domain for standard signals would come up empty and miss the good stuff.
- The OpenAPI spec is real and public but fragmented: each endpoint reference page embeds its own OpenAPI YAML block rather than exposing a single downloadable openapi.json/yaml, so an agent cannot grab the whole contract in one fetch.
- REST examples are multi-language (cURL + @axiomhq/js) and copy-pasteable, but lean on UPPERCASE placeholder tokens (API_TOKEN, DATASET_NAME, AXIOM_DOMAIN) and literal "string" values for startTime/endTime, so an agent must infer real values rather than adapt a realistic sample.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish llms.txt at the apex (axiom.co/llms.txt) or redirect to /docs/llms.txt, and add the docs sitemap to the root sitemap — so an agent probing standard root paths actually finds Axiom's (excellent) agent assets.
- 02Expose a single canonical, downloadable OpenAPI spec (e.g. axiom.co/docs/openapi.json) covering all REST endpoints, in addition to the per-endpoint embedded blocks. This is the biggest single lift to the Discovery score.
- 03Add an AGENTS.md to the primary public repos (axiomhq/axiom-go, axiomhq/axiom-js, axiomhq/mcp) so coding agents get build/usage context in the standard location.
- 04Trim /docs/llms.txt toward the <50KB / ~12K-token target (or split it), and replace literal "string" example values for startTime/endTime with realistic ISO-8601 timestamps to make examples directly runnable.
- 05Consolidate error-recovery guidance: pair the documented status codes (403 auth, 429/rate-limit) with explicit 'what to do next' steps (e.g. wait until the reset epoch, re-scope the token) in one place an agent can find.
Execution coverage · INFORMATIONAL, UNSCORED
Whether an agent can actually complete a call and recover from errors is the deeper Audit layer — documented here, but not part of the Discry Score.
Auth is clearly documented (Bearer API tokens vs. PATs, with the correct header and a note that ingest only supports API tokens). Errors are JSON-encoded and auth failures return 403. Rate limits are well documented via a dedicated API limits page with X-RateLimit-Scope / X-RateLimit-Limit headers, reset windows in UTC epoch seconds, and per-query X-QueryLimit-Limit headers. No cursor/offset pagination or idempotency-key mechanism is documented.