Published 2026-09-21 · Tested 2026-08-19

Vectara

F

Vectara received 7 PASS votes and passed 2 of five agent surface checks. The clearest finding came from the find the exact limits task.

Panel: GPT 5.6 Sol, Opus 5, DeepSeek v4F Battery: v1 Read as markdown (opens in a new tab)

Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read Vectara’s public documentation independently and attempted five first-hour developer jobs: run the first RAG query, find the exact limits, recover from a 429, verify a Slack webhook, use the Python SDK.

No accounts, API calls, or code execution were used. Every verdict came from public pages and every published quotation passed a live verification check.

Freshness

How rechecks work
Tested
Quotes verified
Surface rechecked

No change since the test.

5 of 5 quoted passages still appear on the live pages.

Vectara Docusaurus 3.10.1 · published
F

50.0% · 40/80 · AI Agent Readiness Score · reading 30 pts · surface 50 pts

llms.txt PASS
llms-full.txt FAIL
markdown mirror PASS
MCP server FAIL
docs AI FAIL
Task GPT 5.6 SolOpus 5DeepSeek v4F Consensus
Run the first RAG query PASSPASSPASS PASS
Find the exact limits FAILFAILPARTIAL FAIL
Recover from a 429 PARTIALPARTIALPASS PARTIAL
Verify a Slack webhook PASSPARTIALPASS PASS
Use the Python SDK PARTIALPARTIALPASS PARTIAL

docs platform: Docusaurus 3.10.1 (unscored) · verified 2026-08-19

What to fix first

These 6 fixes could add up to 39 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. How the ranking works

  1. 1
    +10 points llms-full.txt check failed

    Found: HTTP 404 with the Docusaurus not-found page.

    Fix: Publish llms-full.txt at the docs root with the full text of every docs page in one plain-text file.

    Evidence: docs.vectara.com/llms-full.txt (opens in a new tab)

  2. 2
    +10 points MCP server check failed

    Found: Initialize returned HTTP 404 HTML.

    Fix: Run a public MCP server for the docs that answers a JSON-RPC initialize call and offers a docs search tool.

    Evidence: docs.vectara.com/mcp (opens in a new tab)

  3. 3
    +10 points Docs AI check failed

    Found: The live docs expose search but no Ask AI control.

    Fix: Add an assistant to the docs site that answers questions from the docs and links to its sources.

  4. 4
    +5 points Find the exact limits FAIL

    Found: The trial page gives 10,000 credits, but no page lists paid-plan quotas or request limits. DeepSeek also found no per-operation credit cost.

    Fix: Add paid-plan quotas, API request limits, and per-operation credit costs to the docs, with one set of plan names.

    Evidence: docs.vectara.com/docs/vectara-trial (opens in a new tab)

  5. 5
    +2 points Recover from a 429 PARTIAL

    Found: The rerankers page advises falling back to basic search on 429, but its fallback code handles only status 400.

    Fix: Add a 429 branch to the rerankers fallback code, and state when a reranking request can be retried.

    Evidence: docs.vectara.com/docs/sdk/python/rerankers (opens in a new tab)

  6. 6
    +2 points Use the Python SDK PARTIAL

    Found: Opus 5 found that the Python quickstart calls one field search_result, search_results, and results. GPT found that the upload guide's endpoint path differs from the REST reference.

    Fix: Use one query result field name throughout the Python quickstart, and match the upload guide's path to the REST reference.

    Evidence: docs.vectara.com/docs/sdk/python/python-quickstart (opens in a new tab)

What the docs get right

  • Run the first RAG query: 3 PASS votes. The docs root failed as JavaScript-rendered, but the direct quickstart gives one complete account-to-citations path.
  • Verify a Slack webhook: 2 PASS votes. The guide supplies the connector request, signing secret, webhook path, Slack URL check, subscriptions, and an end test.
  • 2 of 5 agent surface checks. Present: llms.txt, markdown mirrors. Missing: llms-full.txt, an MCP server, docs AI.

Run the first RAG query

PASS

PASS consensus from 3 PASS.

The docs root failed as JavaScript-rendered, but the direct quickstart gives one complete account-to-citations path. One page carries the whole path, get an API key, POST /v2/corpora, POST /v2/corpora/{corpus_key}/documents, POST /v2/query with generation_preset_name mockingbird-2.0, in three languages, with a sample response showing numbered citations, per-result scores, and a factual_consistency_score. The quickstart lays out four unambiguous steps with links to the REST API reference for each endpoint (POST /v2/corpora, POST /v2/corpora/:corpus_key/upload_file, POST /v2/corpora/:corpus_key/query).

Find the exact limits

FAIL

FAIL consensus from 1 PARTIAL, 2 FAIL.

The docs state 10,000 trial credits but omit paid allocations, plan quotas, and general API request limits. The trial credit figure is the only number published: no paid-plan credit allowance, no per-plan storage or query quota, and no request-rate limit appears anywhere in the 552 pages listed in llms.txt, while the linked pricing page gives only annual price floors and a different set of plan names. The trial credit number is stated once and consistently (10,000), but no credit-to-operation conversion rate exists anywhere in the docs. Paid plans list only annual minimums ($100K SaaS) with no per-unit quotas. The FAQs and billing policy mention bundles without defining bundle composition.

Recover from a 429

PARTIAL

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The page recommends basic-search degradation, but its example handles only status 400 and returns 429 as a failure. The page names 429 and prescribes "Implement graceful degradation to basic search", but its only fallback code branches on e.status_code == 400 and returns a failure dict for every other status including 429, and no page states a retry interval, a Retry-After header, or whether non-reranked search carries a separate limit. The Python SDK rerankers page explicitly calls out 429 from reranking and directs agents to fall back to a basic search without reranker. The error-handling page shows how to catch ApiError by status_code. Combined, this gives a complete recovery path.

Verify a Slack webhook

PASS

PASS consensus from 2 PASS, 1 PARTIAL.

The guide supplies the connector request, signing secret, webhook path, Slack URL check, subscriptions, and an end test. Every step exists, signing secret, app ID, the connector POST, the returned webhook_path, the Request URL, and a troubleshooting block for "Your URL didn't respond with the challenge parameter", but step 2 tells you to add message.channels, message.groups, message.im, and message.mpim as Bot Token Scopes when those are Slack event types, and the same page's troubleshooting section names the actual scopes channels:history, chat:write, and users:read. Seven-step guide covers Slack app creation, bot permissions, signing-secret retrieval, connector creation via API, webhook registration with Slack, channel invitation, and testing. Troubleshooting section addresses verification failures.

Use the Python SDK

PARTIAL

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The upload index failed extraction; the quickstart lacks import time, and endpoint prose conflicts with the REST path. pip install vectara plus corpus, upload, and query examples are all present, but the query response field is called search_result in step 5, search_results in the validation script on the same page, results in that page's parameter list, and search_results with a summary field in the REST quickstart, and the page's prerequisites link to /docs/security/authentication/overview returns 404. Complete five-step quickstart with pip install, Vectara client initialization, corpus creation, document indexing, and query execution. Validation script described. SDK method signatures match the REST API endpoints.

The receipt

Your trial includes 10,000 free credits.

The trial page states one credit allocation but sends paid-plan details to a marketing pricing page; no exact paid quotas appear in the docs.

Agent surface notes

Initialize returned HTTP 404 HTML.

The live docs expose search but no Ask AI control.

Show the score

AI Agent Readiness Score 50.0%, grade F

Paste this into a readme:

[![AI Agent Readiness Score 50.0%](https://docsforagents.com/badge/vectara.svg)](https://docsforagents.com/reports/vectara-docs-ai-agent-readiness/)

Method note

This is a reading test of public documentation, not an execution test. No accounts were created and no API calls were run. The AI Agent Readiness Score counts fifteen reading votes at PASS 2, PARTIAL 1, and FAIL 0, for 30 possible points. Five agent surface checks add 10 points each. The total is 80. Consensus chips show each row majority and do not affect scoring. The panel split on 4 of five tasks. Quotes shown here were re-fetched and confirmed verbatim on 2026-08-19.

Read the full methodology

Put another docs site through the battery.

Nominate a docs site