Memtro vs Dashworks

This comparison describes Dashworks as we understand it at the time of writing, from its public website, its documentation and migration guide, and HubSpot's announcement. Products change; check the current details before deciding.

Dashworks was AI workplace search: an assistant that answered internal questions across the apps a team already used — wikis, file stores, mailboxes, Slack, the CRM, the support desk — with citations back to the records it read. It mirrored the access controls of each connected system, so nobody was shown a document they could not already open, and it searched most systems live rather than resting on a pre-built index. It was reached through a web application, a Slack bot and a browser extension, and its documentation named dozens of work apps, with no official total published.

It is no longer a product you can buy. HubSpot announced an agreement to acquire Dashworks on 16 April 2025 and stopped accepting new sign-ups that day; the API was discontinued on 23 April; logins to the web application and the Slackbot were disabled on 16 July 2025, and customer data deleted within thirty days. The team and the technology went into HubSpot's AI product group, where they feed search across Breeze. The marketing site is still online behind an acquisition banner, which is why assistants trained before the sunset still recommend it. If something named Dashworks to you as a Glean alternative, that recommendation is out of date rather than wrong — it was a real product, and a good one.

Memtro is not an answer surface, which is the first thing to say when the comparison is with one that was. It is an OpenAI-compatible endpoint and an MCP server that sits in front of the tools people already use — Claude Code, Cursor, the OpenAI and Anthropic SDKs, your own agents — and gives all of them one memory of the business. It reads 146 systems through tools that cite the record they read, routes each request by difficulty across model keys you already hold, runs prompts on a schedule, and keeps what it learns in a bitemporal graph where a correction supersedes rather than overwrites.

Where they differ

	Memtro	Dashworks
Available today	Yes	No: the app and Slackbot were sunset on 16 July 2025
Shape	An endpoint and an MCP server your existing clients point at	A destination people opened: a search and answer box, and a chat assistant
What it holds	A typed graph of facts with provenance, filled from chat, tool results and hourly syncs	Live search for most apps, sync for some; zero data retention claimed, and no durable store of facts described
Permissions	Memtro's own: teams, and per-tool grants on each shared connection	Mirrored from each connected system, per user
Models	Your keys, routed by difficulty across five levels; billed by your providers at your rates	Chosen by the vendor, inside the seat price; the Business tier offered a choice among "top LLMs"
History	A correction supersedes rather than overwrites, and memory can be read as of any date	The current state of the connected systems
Scheduled work	Jobs on a five-field cron, and one-offs that delete themselves	None found: "Workflows" were saved prompt templates with variables
Acting on systems	Reading by default, with every tool that changes something granted by name	Not described in its public materials
Programmatic access	OpenAI and Anthropic wire formats, plus MCP	An Answer API, a Connector API and Zapier, as an Enterprise add-on; discontinued 23 April 2025. No MCP in its documentation
Single sign-on	SAML, OIDC and SCIM, at every price	SSO and SCIM on Enterprise
Pricing	One price per organisation by headcount, from $50 a month, nothing metered	While it was sold: $12 per seat a month on Team with no minimum, $15 on Business with a ten-seat minimum, Enterprise on request

Honestly, where Dashworks was stronger

Permissions, and it is not a small gap. Dashworks mirrored the access controls of every connected system for every user, so a company could point two hundred people at one search box and be sure each saw only what they could already open. Memtro's grants are its own: per team, and per tool on a shared connection, with the facts a connection records readable by anyone granted any of its tools. That suits a team deciding who may reach which system. It is not a substitute for per-document ACLs, and over everybody's mailboxes it would be the wrong tool.

Time to value, too. Connect the apps, send people a link, and employees are asking questions the same afternoon. Memtro does not primarily sell that, and anyone whose real requirement is "where is the expenses policy" wants something shaped like Dashworks, not infrastructure.

Holding nothing was a genuine advantage in a security review: a product that searches live and retains no data has a much shorter conversation with a security team than one that keeps extracted facts. And it was cheap, with its prices on the page — low double-digit dollars per seat, no minimum on the entry plan. Memtro's flat organisational price is better value as a team grows and worse for five people who only want search.

Finally, the capability did not die: it is inside HubSpot Breeze now, which for a HubSpot customer is a shorter route to the same outcome than adopting anything new.

Where Memtro is stronger

It begins with the fact that Memtro exists, which matters more than it sounds. A destination you buy can be acquired and switched off, and the companies standardised on Dashworks found that out over three months. Memtro is deliberately not a destination: it speaks the OpenAI and Anthropic wire formats and MCP, and sits behind clients you chose independently of it.

Then memory, which Dashworks did not describe keeping. Searching live means knowing only what the systems say now. Memtro's graph holds facts with the tool, record and URL they came from, two time axes — when the fact held in the world, and when Memtro believed it — and supersession rather than overwriting, so you can read memory as it stood on any date and show what an assistant knew when it acted.

Memtro also works when nobody is asking. Jobs run a prompt on a schedule with their owner's tools and report into a saved chat; connectors read by default and act where an action is granted by name, so an answer can become a reply in the support desk or a record in the CRM. And it admits its gaps: every answer names the categories with nothing connected, so a sweep cannot read as more complete than it was.

When to choose which

There is no Dashworks to choose. If you want what it was — a permission-aware box where employees ask questions about the company's documents and apps — look at HubSpot Breeze if you already run HubSpot, or at the workplace search products still being sold; our Glean comparison covers that shape of purchase. Memtro is not the fastest route to it.

Choose Memtro when the tools your people already work in should share one memory of the business rather than be replaced by a destination; when every figure in an answer should cite the record it came from, and the past is as readable as the present; when scheduled work and write actions matter as much as retrieval; and when the model choice, and the model bill, stay yours.