Published 2026-09-22 · Tested 2026-09-22
LiveKit
A+LiveKit received 13 PASS votes and passed 5 of five agent surface checks. The clearest finding came from the find the exact limits task.
Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read LiveKit’s public documentation independently and attempted five first-hour developer jobs: run the first voice agent, find the exact limits, recover from a session error, verify a webhook, use the Python Agents 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. Preflight substitutions: T3: An AgentSession error event replaces HTTP 429, because LiveKit documents quotas and rate limits without naming a 429 response.
Freshness
How rechecks work- Category
- Voice & speech
- Tested
- Quotes verified
- Surface rechecked
- Not yet rechecked
97.5% · 78/80 · 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 |
|---|---|---|---|---|
| Run the first voice agent | PASS | PASS | PASS | PASS |
| Find the exact limits | PARTIAL | PASS | PASS | PASS |
| Recover from a session error | PASS | PASS | PASS | PASS |
| Verify a webhook | PASS | PASS | PASS | PASS |
| Use the Python Agents SDK | PASS | PARTIAL | PASS | PASS |
docs platform: Custom Next.js (unscored) · verified 2026-09-22
What to fix first
The docs passed every reading task and every agent surface check.
What the docs get right
- Run the first voice agent: 3 PASS votes. One clear path via lk agent init my-agent, lk agent dev, and the Agent Console. CLI commands reside in JS-rendered code blocks but are present in the page source.
- Recover from a session error: 3 PASS votes. Two complementary mechanisms documented: automatic fallback adapters with backup providers, and manual error handling via the error event's recoverable field. Includes pre-recorded audio fallback when TTS fails.
- 5 of 5 agent surface checks. Present: llms.txt, llms-full.txt, markdown mirrors, an MCP server, docs AI.
Run the first voice agent
PASSPASS consensus from 3 PASS.
One page covers Cloud signup, CLI setup, project creation, dependency installation, local execution, and browser microphone testing. One numbered path runs end to end: install the CLI, lk cloud auth, lk agent init my-agent --template agent-starter-python, uv sync, lk agent dev, then open the printed Console link and click Start a session, with the full agent.py printed on the same page. One clear path via lk agent init my-agent, lk agent dev, and the Agent Console. CLI commands reside in JS-rendered code blocks but are present in the page source.
Find the exact limits
PASSPASS consensus from 2 PASS, 1 PARTIAL.
The Server API rate is exact, but paid limits require the pricing page and the two pages disagree on the Enterprise custom-voice limit. A single Quotas and limits page carries the Server API rate, the Build-plan concurrency caps, LLM limits of 100 requests and 600,000 tokens per minute, and the metered allowances, and the RoomService API reference states no competing number. All quotas and rate limits documented on a single page with consistent values across concurrency, API, LLM, STT, TTS, and observability categories.
Recover from a session error
PASSPASS consensus from 3 PASS.
The events page distinguishes recoverable failures and supplies complete Python handlers for LLM, TTS, realtime, and STT recovery. The events page defines recoverable versus unrecoverable errors with runnable handlers and a pre-recorded audio path for TTS failure, and links to Fallback strategies, which supplies the complete backup-provider correction through inference.STT(fallback=[...]) and stt.FallbackAdapter. Two complementary mechanisms documented: automatic fallback adapters with backup providers, and manual error handling via the error event's recoverable field. Includes pre-recorded audio fallback when TTS fails.
Verify a webhook
PASSPASS consensus from 3 PASS.
One page covers Cloud configuration, a test event, POST receipt, raw-body handling, and JWT validation with WebhookReceiver. One page covers Cloud dashboard setup with a signing API key, a Send a test event control, the application/webhook+json content type, the raw-body requirement, and WebhookReceiver verification code for Node.js, Go, and Java; no Python receiver example is given. End-to-end documented flow: Cloud dashboard configuration, server-side WebhookReceiver class with token verification and payload hash matching, and complete Python reference code.
Use the Python Agents SDK
PASSPASS consensus from 2 PASS, 1 PARTIAL.
The Python reference identifies the SDK, and the quickstart's complete agent.py uses the same Agent, AgentSession, and start API. The Python SDK is identified and the quickstart's agent.py matches the AgentSession and inference API used on the events and fallback pages, but the Reference index link https://docs.livekit.io/reference/python/livekit/api.md returns HTTP 404, the agents reference has no Markdown mirror, and no page states a pip install or uv add line for livekit-agents outside the CLI template. Official SDK at github.com/livekit/agents with pip install. README and API reference at docs.livekit.io/reference/python/livekit/agents/ consistently show the same AgentSession constructor signature.
The receipt
All projects have a Server API rate limit of 1,000 requests per minute.
All quotas and rate limits documented on a single page with consistent values across concurrency, API, LLM, STT, TTS, and observability categories.
Agent surface notes
Initialize returned JSON-RPC protocol 2025-06-18 and serverInfo name livekit-docs version 1.5.0.
The hydrated docs DOM exposes an Ask AI control with aria-labels "Ask AI" and "Ask AI chat".
Show the score
Paste this into a readme:
[](https://docsforagents.com/reports/livekit-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 2026-09-22.
Put another docs site through the battery.
Nominate a docs site