Launching today
DocsAlot
Documentation that works for both humans and AI systems
297 followers
Documentation that works for both humans and AI systems
297 followers
DocsAlot turns scattered help center articles, knowledge base, and developer docs into one source of truth for humans and AI agents. It includes hosted MCP, llms.txt, and skill.md. Your docs show up in AI answers, onboarding gets faster, and agents stop reading stale context.
This feels very relevant right now.
Docs used to be "just" onboarding and support, but now they also decide what AI tools and agents understand about your product. If the docs are stale, the agent context is stale too.
We’re working on our own help center and llms.txt setup for our product, so I really like the idea of treating documentation as a maintained knowledge layer, not a side project someone updates when they remember :)
Curious how DocsAlot handles product changes over time. does it detect when docs are outdated from changelogs/GitHub/product updates, or is the maintenance workflow more manual right now?
DocsAlot
@andrasczeizel thanks for the kind words. Yes, it does in fact detect outdated docs from source-code, and going to the dashboard. and recommends updates, for approval.
It even monitors its own traffic, and searches and recommends updates based on what users are querying.
happy to help you guys out for free, if you book a call on my calendar. We have some interesting agentic data on how agents traverse help-centers.
Faizan, this lands at the right time :) But my honest first thought is the one you'll probably hear a lot: llms.txt, skill.md and hosted MCP are becoming checkboxes, Mintlify and GitBook are already bolting them on. Emitting an AI-readable format won't stay a differentiator for long.
The line that actually caught my eye is the one you dropped to Andras: "data on how agents traverse help-centers." That feels like the real moat, using how agents actually read docs to restructure the content for them, not just expose it in a format everyone will have in 6 months. Is that where you're heading (scoring and reshaping docs for agent comprehension), or is the core bet still the unified output layer?
Congrats on the launch! ;)
DocsAlot
@keirodev thanks for the msg. Yes optimizing the docs for agent adoption is the primary objective of docsalot. and we do indeed capture a ton of data and surface them to our customers. Docsalot also drafts articles based on that traffic, and gives suggests on how to structure the docs for better consumptions with less tokens.
I do want to say that docs is just the start, we have bigger things in mind :)
the "detect outdated docs from source-code" part is the piece i'd want to stress test before trusting it on a real repo. false negatives are one thing, but false positives (flagging a doc as stale when the underlying behavior didn't actually change) seem like the bigger risk since that's what erodes trust in the tool and gets people ignoring the suggestions after a few bad flags. how does it decide a doc is stale, does it diff behavior or just correlate with commit/PR activity touching related files?
DocsAlot
@omri_ben_shoham1 mostlly looking at recent commits, and diffs. and then deciding what to focus on. We maintain our own custom diff for the docs, which helps with not loosing too much context.
that lines up with what I expected honestly. the risk i'd still watch for is a big refactor PR that touches a file's structure but not its actual behavior - that could trigger a stale flag even though nothing user facing changed. does the custom diff try to separate structural churn from behavior changing edits, or is any touch to a related file weighted the same right now
"Your docs show up in AI answers" depends entirely on how the underlying models are trained and updated, which DocsAlot doesn't control. The llms.txt standard is still not universally respected across major models and crawlers. What's the realistic expectation for how quickly and consistently docs actually surface in AI answers after setting this up, or is that more of a long-term bet on the standard gaining broader adoption?
DocsAlot
@ansari_adin We constantly probe the AI platforms (codex, claude, cursor even and claude code), and do a benchmark with base line prompts regularly.
The responses tells us whether your product is being recommended or not, and what we need to do to make it consistently recommended
As for realistic expectation, its similar to SEO, can take anywhere from 1-3 months, depending on your prior brand value.
But we can give you observability.
The failure mode I keep hitting wiring agent docs through MCP: the agent grabs a confident, plausible snippet that's one version stale and never flags it, because retrieval has no notion of 'this section is old.' Emitting llms.txt is the easy part. The valuable bit is a freshness signal inside the MCP response itself, so an agent can tell a current param from a deprecated one. Does your outdated-doc detection surface a staleness marker in the MCP payload, or only in the dashboard?
DocsAlot
@dipankar_sarkar So we serve the docs in real time, via manifest fetch, and then specific documentation fetch. This means if the docs are fresh the repsonses are fresh.
Now how to keep the docs fresh is the main problem. Which docsalot solves, by synching your inputs (source-code, website, slack, discord, intercom), with the actual knowledge-base on autopilot.
Finally, someone is tackling this! Writing docs that LLMs can easily parse while keeping them readable for human developers is such a tricky balance right now. Does this integrate directly with GitHub repos to keep the docs synced with the codebase?
DocsAlot
@doganakbulut Thanks for the comment. Yes it does, via our github app, that automaitcally tracks more than one repositories, in a many-to-one fashion.
This feels timely. Docs are no longer just for users and support teams. AI agents also need a reliable source of truth now.
Curious how DocsAlot handles drift over time. If the product changes, does it detect outdated docs automatically, or do teams still need to manually trigger updates?
DocsAlot
@abhishekpatel you can set triggers based on events or fixed schedule.