Published 2026-09-21 · Tested 2026-08-19
Reducto
AReducto received 11 PASS votes and passed 5 of five agent surface checks. The clearest finding came from the recover from throttling task.
Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read Reducto’s public documentation independently and attempted five first-hour developer jobs: parse a document, find the exact limits, recover from throttling, verify an async 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- Category
- Data pipelines & documents
- Tested
- Quotes verified
- Surface rechecked
No change since the test.
5 of 5 quoted passages still appear on the live pages.
95.0% · 76/80 · AI Agent Readiness Score · reading 30 pts · surface 50 pts
Agent surface checks · 50/50
The Reading Test
| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
|---|---|---|---|---|
| Parse a document | PASS | PASS | PASS | PASS |
| Find the exact limits | PASS | PASS | PASS | PASS |
| Recover from throttling | PARTIAL | PARTIAL | PASS | PARTIAL |
| Verify an async webhook | PASS | PASS | PASS | PASS |
| Use the Python SDK | PARTIAL | PARTIAL | PASS | PARTIAL |
docs platform: Mintlify (unscored) · verified 2026-08-19
What to fix first
These 2 fixes could add up to 4 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. How the ranking works
- 1
Found: The agent guide defines 429 as too many concurrent requests, while Rate Limits says the limits cap requests per second.
Fix: Rewrite the agent guide's 429 entry to match the Rate Limits page, and link that page's symptom-to-action table.
- 2
Found: The quickstart says the SDK requires Python 3.8+, while the Python SDK overview says 3.9+.
Fix: State one minimum Python version and use it on both the quickstart and the SDK overview.
Evidence: docs.reducto.ai/
sdk/ python/ overview (opens in a new tab)
What the docs get right
- Parse a document: 3 PASS votes. The quickstart covers account creation, key setup, SDK installation, local upload, synchronous parse, and result access in one page.
- Find the exact limits: 3 PASS votes. The linked pages state 1,000 and 200 req/s limits, plus US/EU baselines of 200/60, 350/120, and 500+/275+.
- 5 of 5 agent surface checks. Present: llms.txt, llms-full.txt, markdown mirrors, an MCP server, docs AI.
Parse a document
PASSPASS consensus from 3 PASS.
The quickstart covers account creation, key setup, SDK installation, local upload, synchronous parse, and result access in one page. The quickstart runs account signup, key creation, environment variable, SDK install, upload, parse call, and a field-by-field response table in one unbroken sequence, with a separate top-of-page shortcut path labeled for coding agents. Every step from API key creation to structured JSON output is documented in a single page with working code in four language ecosystems.
Find the exact limits
PASSPASS consensus from 3 PASS.
The linked pages state 1,000 and 200 req/s limits, plus US/EU baselines of 200/60, 350/120, and 500+/275+. Edge limits are stated exactly per API key (1,000 req/s all endpoints, 200 req/s on GET /job/{job_id}) on the Rate Limits page, and the Concurrency Throttle page gives one table with side-by-side US and EU baselines (Standard 200/60, Growth 350/120, Enterprise 500+/275+ custom) in concurrent batches, with no conflicting numbers anywhere else in the corpus. Edge rate limits (1,000 req/s all endpoints, 200 req/s for GET /job) and per-tier concurrency baselines are stated in a single page each, with US and EU values in the same table.
Recover from throttling
PARTIALPARTIAL consensus from 1 PASS, 2 PARTIAL.
The rate-limit page maps each 429 body to a fix, but the agent guide incorrectly describes 429 as concurrent-request throttling. The Rate Limits page diagnoses the condition precisely with distinct 429 body codes ([CODE 1000], [CODE 2000], and bodyless infra-shed 429s) plus a symptom-to-action table, but two other pages contradict it: the agent guide defines 429 as "Too many concurrent requests" when concurrency pressure returns 200 after a queue wait, and the Error Codes page claims async endpoints "have higher limits" when the edge cap is stated as applying to all endpoints. The How to Tell Which Limit You're Hitting table maps each 429 variant to its cause and action; the Error Codes page contributes code-level handling examples with SDK exception types.
Verify an async webhook
PASSPASS consensus from 3 PASS.
One page covers dashboard access, endpoint creation, async.update delivery, signing-secret retrieval, and three-header verification. One page covers the whole loop: reaching the Svix dashboard through Studio or POST /configure_webhook, adding the endpoint, submitting with async.webhook.mode set to svix, the exact payload shape, and runnable Python and TypeScript verification using the svix library with all three svix-id, svix-timestamp, and svix-signature headers. A single page covers dashboard access, endpoint configuration, job submission with webhook mode, payload format, and production-ready signature verification code in Python and TypeScript.
Use the Python SDK
PARTIALPARTIAL consensus from 1 PASS, 2 PARTIAL.
The SDK form matches REST, but the quickstart requires Python 3.8+ while the SDK page requires 3.9+. The package name, import path, and call form (pip install reductoai, from reducto import Reducto, client.parse.run(input=...)) match across the quickstart, the SDK section, and the agent guide, but the quickstart states the requirement as "Requires Python 3.8+." while the SDK overview states 3.9+, so the reader must guess which floor is real. The official SDK (PyPI package reductoai) has its own documentation section with a minimal upload-and-parse example that matches the API form used in the Quickstart and the REST API reference.
The receipt
Too many concurrent requests. Retry with backoff.
The rate-limit page maps each 429 body to a fix, but the agent guide incorrectly describes 429 as concurrent-request throttling.
Agent surface notes
Initialize returned JSON-RPC protocol 2025-06-18 and server name Reducto.
The live overview exposes an Ask Assistant control.
Show the score
Paste this into a readme:
[](https://docsforagents.com/reports/reducto-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 2 of five tasks. Quotes shown here were re-fetched and confirmed verbatim on 2026-08-19.
Put another docs site through the battery.
Nominate a docs site