Subscribe
Share
Share
Embed
Podcast host Mike talks to Tessa and Steve from Tabstack about building an API that gives standalone models access to the web.
- Tessa - https://x.com/tessak22
- Steve - https://x.com/JustSteveKing
- Tabstack - https://tabstack.ai/
- Steve's PHP Library for Tabstack - https://github.com/JustSteveKing/tabstack
- Mozilla - https://www.mozilla.org/en-US/
Creators and Guests
Host
Guest
What is APIs You Won't Hate?
A no-nonsense (well, some-nonsense) podcast about API design & development, new features in the world of HTTP, service-orientated architecture, microservices, and probably bikes.
Hello, and welcome to APIs You Won't Hate.
My name is Mike Bifulco, I am sitting
down for a long overdue chat with
two of my friends from the internet.
I don't think I've ever been in the
same room with either of you, but I'm
100% confident I've been in the room with
people who've been in the room with you.
That's basically like being best
friends as far as I'm concerned.
Today I'm hanging out and chatting
with Tessa Kriesel and Steve
MacDougall both from TabStack.
Tessa is the technical founding
developer go-to-market at TabStack, and
Steve is a developer experience works
in developer experience at TabStack.
Why don't we start there?
Tessa, why don't you give a,
a brief intro of yourself.
G- give me the the elevator
pitch, the 30 seconds on Tessa.
Oh gosh.
Tess is wild and weird
and chaotic and crazy.
No like engineer by trade, really
love just like building things.
Got pulled into go-to-market about
10 years ago, which started in dev
rel and it just unfolded from there.
And I would say most recently becoming
somewhat of an AI whiz, really learning
all the tools, all the things, getting
back into engineering big time for sure.
And just really enjoying this like
kind of chaotic world that we're
living in where it's like, oh, I'm
like engineering my go-to-market and
building things in a matter of hours.
And yeah, that's a good summary of me.
Yeah, right on.
I did forget part-time dairy farmer.
Oh, that's cool.
I'll have more questions
about that without a doubt.
Steve, how-- welcome as well.
How about you?
What's what's been going on with you?
And tell us the intro on Steve.
I think, so.
Yeah, that makes you and I star-crossed
kindred opposites, 'cause I'm the
API guy who really likes the front
end part of things and yeah we're
like linked for better or worse
to one another's work, I think.
Hey, look, everyone has their expertise.
I'm here just to do what I'm
good at, all right, so you're
both working with TabStack.
Can you tell me about TabStack?
For folks who haven't heard of
TabStack before, what is it?
So TabStack is what I think of as like
a amazing Swiss Army knife, like in
your tool belt, in your agentic world.
And you can obviously use it
outside of, an agentic system,
but TabStack essentially gives
you access to the internet.
And I know that sounds silly, but
like in the agentic space, right?
Claude and ChatGPT and all
the frontier models have this
beautiful harness that's trained.
They know how to access the web.
They know how to get data.
They know how to cite
research calls, right?
They know how to go out and do things.
But all these other models that
we wanna work with, maybe we don't
want the frontiers or maybe they're
gonna become unaffordable very soon.
I just saw someone get kicked off of a
$200 plan on like ChatGPT, and I was like,
"Oh, this is weird." So we're all gonna be
thinking about we need these other models.
We need models that we control, dive into
the open source side of things, right?
It's just gonna be the
evolution of the ecosystem.
And so as you do that, you're gonna have
to train those models on how to scrape
the web, how to do research calls, how to
go do things, how to get things, right?
Give them browser infrastructure.
And so TabStack has essentially three
very distinct endpoints, but about five
different features that allows you to
essentially go out and either scrape
data make a research call to get a cited,
get cited information back on something
you're looking for research info on,
or send it out to go do a task, right?
"Hey, can you go on the web and fill
out this form for me? There's five
forms, go grab my data and go do this
task and come back." And so that,
that's the automate endpoint, and I
actually use that one quite a bit.
So that's kind of TabStack in a
nutshell, of just really being this
like beautiful Swiss Army knife to
give your agent or your application,
if they're working with LLMs, access
to the web if they don't have it.
I think those are the things that
as soon as you start doing things,
touching agents and start to dream
about, "Hey what can I make this thing
do?" It starts to reach out into the
web and touch more and more things.
But there's a lot of questions
about that as a developer type,
and there are many things as a…
Gosh, I'm trying to include the voice of
the, the newly technical non-developer
who is making things, who just have maybe
some assumption that "Of course my robot
ally can use the web." And there's a
whole lot that comes along with that, and
I'm eager to ask you a lot about this.
But before I go any further, the thing
I wanna say, too, is that when Steve
reached out about this interview, he
told me "Hey, I have something I wanna
talk about," and he gave me the pitch for
it, and I basically said, "Say no more.
I don't wanna know anything about
this until you tell me about
the project as we're talking."
And I'm looking up the exact direct
quote that he sent to me was "We're
working on w- on a product at the moment
with Mozilla that is basically API
access to the web." And of course, with
our squad and API developers, that's
those are the keywords I need to hear.
What on earth does that mean?
What are we talking about today?
Yeah
So one thing that I built when I first
joined Tapstack is I built out this
competitive intelligence tool that
essentially is only powered by Tapstack.
And so what it does is it goes out
and it keeps a radar on every single
one of our competitors, and it shows
like what's a recent blog, like all
their founding information, if their
messaging changed, if their jobs board
changed, like just finite details.
And so when I started at Tapstock,
A, it was my way to learn the API
and build with it, but when I joined
I was like, "Haha, I've got all my
information on my competitors," right?
And it's, that's like a tool
that a go-to-market person
wants right next to them.
Then I would say on the contrary
end, like we've talked about,
I don't know, Steve, d- did we
ever build the reverse job board?
We talked about doing a reverse job board.
Yes
When you break down the different
kind of like endpoints of what you can
do really, like the extract endpoint
is very JSON driven, and so it's
got that schema defined around it.
And so when you think about like
web scraping in general, you're
constantly having to like map to
IDs or map to wherever, right?
Or oh, here's the identifier.
Where this is like schema based and
there's intelligence behind it, and so it
can go to any website and you're not gonna
have a broken scraper the next time you
use it because it's gonna be able to adapt
and evolve and figure out what it needs.
I think one of the first questions
I have is it sounds like there's a
handful of things that are the core use
cases or maybe like the base endpoints.
I've heard scrape and
research and automate.
What is the balance of those things
being deterministic and like an API
that will always return the same
versus agentic and something that
is gonna go kick off a the black box
workflow that an LLM would run through?
Sure
the citations are arguably like,
I guess maybe deterministic.
Like we were just working with somebody
where he was consistently looking for
the same type of data, but we were
gonna use a research endpoint to go
grab that data and then validate via the
citations URL to make sure that the data
was like, oh yeah, that's exactly what
we're looking for each and every time.
But that is definitely a little bit
more open-ended on the research endpoint
We actually run a, like a subset of a
whole bunch of different kinds of evals
that happen internally behind the scenes.
And so what we're doing is
consistently evaluating whether
or not people are getting some of
those deterministic results, right?
We're not consistently gonna
always be the same because we're
still surfacing the internet.
It's the live internet that it's
going out and actually working with,
and the research stand-standpoint,
it also has a live corpus of data
that is also not live, right?
And so it can compare both.
And so anyways, as we're running
those evals, we try to think
about those things, right?
What kind of results are we
getting, and how do we make sure
that the API is performing at peak?
Because when you are working with
an LLM, it's… You don't know
what you're gonna get, right?
And so we're still trying to as
humans, figure out how to put those
little harnesses or those kind
of barriers around these LLMs.
And so we feel like internally
we've tried really hard to do that.
Let's start here then.
If I'm signing up t- for Tapstack
for the first time, what does
"Hello, world" look like?
Ah, cool
Right
And so like from the jump, you're
able to test things out and
start using the functionality and
probe around with it?
And so the API part of this then,
if it sounds like I can in-invoke
these these things programmatically
in some sense too tell me about that.
Is that tool calls with an
MCP in addition to like SDKs?
How does that all work?
Hey
Yep
Yeah, that's a workflow I think we're
all getting used to as builders of things
is like we not very long ago lived in
a world where it was do something, it
comes back, do something, it comes back.
And now our UX being interrupted in
that sense is a really interesting
case, and it's turns out like far
more natural in a lot of ways.
But as a I'm used to telling the
computer to go add two numbers and
it gives me those numbers back,
and then I do the next thing.
It's a very different experience.
Tessa, I'm curious, as you're approaching
this, what does it look like for you?
Like what are the types of people that
you're identifying as developers who
should be taking a look at this or
like who have interesting use cases?
How are you identifying and finding
the types of people that should be
diving in and giving Tab Stack a go?
So we actually are, like, in
the middle of our reposition.
If you go to Tapstack today as we're
recording this, you're gonna see our
old positioning, but if we go-- If you
go there the day that this is live,
it'll likely be our new positioning.
And so we're, like, right in the middle
of that pivot, and so actually to speak
to that, 'cause I think it, it makes
really good content, honestly, to share
these testimonials and s- true stories
is that we had Tapstack launched as
a, essentially like a, a similar to
how we're talking about it, right?
But we weren't seeing a ton of
actual adoption and retention.
And so it was like, okay,
what does that mean, right?
That means we're not
solving the right problem.
And yes, we were solving some
problems, talked to as many users
as we could and all the things, and
just started to realize that where
Tapstack really shines is where The
web is really hard to access, right?
And like I was saying earlier with
Claude and ChatGPT, it's easy to access
that way, but not so easy to access if
you're picking up an open model or doing
anything that's non-frontier, right?
And so with that reposition, it's been
shifting over to you're working with
your own models, and that's the A number
one like identifier for our ICP, is that
they're working in their own models.
And it's not that you can't use Tabstack
if you're not working in your own models.
I use it every single day in my
Hermes agent, and I'm a Claude
and Codex and GPT user, right?
But it's just that the pain is so
annoyingly painful when you're using a
local model because you don't realize,
like you said, that you don't have
access to the internet inside of
these very core base level models.
And so that's like the A1
info- identifier, right?
Is are you working in that local model?
But then it's for the most
part building agentic systems.
So it's not that, again, you can't
use Tabstack and bring it inside
of an app and have no, artificial
intelligence in there at all.
You absolutely could.
Tabstack is all managed, and so
the artificial intelligence is
all managed inside of that call.
So we go make that intelligence call,
come back, make sure you get the right
result, and feed that back to you.
So you could hypothetically use
it in any kind of app instance.
But again, not what we're looking for.
We're really looking for that dev
that's working with those local models,
that's bringing in this agentic system.
Maybe they're working with an agent,
maybe their app is working in this way
and they just need access to the web.
And they need… Ideally, what I'm
thinking is the rest of their stack
is probably pretty secure because
Tabstack is managed, and so that
call happens inside of Tabstack.
It goes out, it reaches out to the LLM,
it uses the browsers it needs to, it
does what it needs to do, and it comes
back and it gives finished output.
So you're not dealing with, "Oh, hey,
I've got a search API that's gonna go give
me a whole list of results, and then I
still have to decipher all those results."
It's very much about, "No. Hey, give
me this finished output so we can feed
it back in." And so Mozilla's stance of
being very trust-driven, very privacy
by nature really resonates there.
And I think with the Mozilla mission,
that's really the direction that
we're going just because we- For a
whole multitude of reasons, but I have
this hunch that like these frontier
models are gonna get overpriced.
We're gonna get pushed out of them.
Some of them will maybe get to use
them in work, but in a personal
level, we won't have access to them.
But also we're just
gonna want more freedom.
And so I'm just seeing that
ecosystem is gonna build up.
And so the nice thing is like now I get
to go position ourselves in a fairly small
total addressable market where we can
learn the players, learn the ecosystem,
get involved in local meetups, get
involved in sort of newsletters, right?
Because the local model space is
actually fairly it's fairly young, right?
In general, we're seeing some fairly
young technology and AI across the board,
but I think what's really cool is being
able to get into an ecosystem that just
so beautifully aligns with Mozilla's
like values and what they go after.
So that's a whole long thing, but it is
I guess one little TLDR Bo is it's like
stupid simple to honestly install and
if you're comfortable in any kind of a
agent situation, even if you're using
Claude and you're a go-to-market person
and you don't code, it's the install is
so like easy and it's so powerful that
it can very well be for other cases.
But I do think that it's gonna be, fairly
technical that are working and building
in these, these kind of newer ecosystems.
So
Yeah, it sounds like the ability to
wrap an, a existing model with the
sticky bits that let it pull things
from the internet with a predict- in
a predictable way is something that
I guess those, like the great big
companies are able to do because they
can hire, armies of engineers to do that.
But if you're using a smaller model
or developing your own, like that
surface area isn't necessarily there.
Yep,
exactly
Yeah.
Let's talk about the costs a little bit.
So you mentioned on signup there's
some free credits and whatnot.
What is it actually I
guess how is it metered?
Like Cost structure look like?
Yeah.
So each of the different calls have
a different credit criteria to them.
So like for example, a
research call is 250 credits.
And so for context, when you join,
the free tier gets 10,000 credits.
So if you do all the math there,
you can actually run quite a bit of
free calls with that credit balance.
I think extract runs anywhere
from 50 to 100, depending on how
you're asking for the shape back.
Automate I believe is a little bit higher.
I wanna say at 350.
Somewhere in that range.
And then you get that, those 10,000
free credits, and you get those every
single month, so you do get to keep
kinda piling on with those free credits.
I think we do actually have a lot of
users that are like, "The free credit
tier is good for me," 'cause they're
just adding it onto their agent,
their Hermes or their OpenClaw agent.
And it's just like a great additive
tool, and they don't use it a ton.
They're just making a few research
calls every once in a while.
But we do have different tiers, right?
And so there's like a $10 sort of personal
level, and then we kinda go up from there.
Like a $100 level, that's more for okay,
yeah, you're making a lot more calls.
You're doing things
that are more reliable.
Like I would say my competitive intel
tool, I don't know what my usage is, but
I'm definitely the top user of TabStack.
And I think my competitive intel
tool runs quite a few calls.
So I would say like just thinking
about like you're building a
full-fledged app, you're probably
looking at that kind of $100 range.
And then there's a $500 plan for a team
that's kinda using it across the board.
And I would say that's actually where
I'm trying to position us is can we
come in and try to actually address
teams that are working with local models
that all want that kind of web access
that's added onto that team model that
they have, we shall see though, right?
The world tells us when
you put marketing into it.
Sure
Yeah.
Marketer's nightmare, honestly.
there's a really cool organization inside
of Mozilla it's called New Products.
And so what New Products is intended
to do is to build new products.
But the goal of that organization
is to actually drive revenue back
into Mozilla, the foundation,
and the organization as a whole.
So for anyone who doesn't know any- that,
know about Mozilla it is a non-profit
foundation backed organization.
And so Mozilla has been around
for a very long time, just
stewarding good web, really caring
about trust and data privacy.
And so they're the builders of
Firefox, if you're familiar.
And so how Mozilla plays into that
is that Tab Stack is actually a
product inside of New Products.
And so it's just one of the sort of
different kinds of bets that we have
going right now that we think could
potentially drive revenue back to Mozilla.
We shall see, right?
I think-- I joke that Tab Stack came a
year too late because it would've been
perfect to add onto Claude a year ago
when it couldn't access the web, right?
Or in today's day, it's a year too early.
I think that local models are gonna
be quite popular a year from now.
And so it's a bet that I'm hoping
that we can keep making at Mozilla to
play into this ecosystem, alongside
of a couple other new products that
we're working on the same team.
Yeah.
Ooh, our team is actually quite small.
So in terms of full-time folks,
there's five, six of us now.
Six of us.
So Steve is on the team part-time,
although I feel like once we've got all
these products built, Steve is gonna be,
like, a full-time thing where we're gonna
need someone who's dedicated to this.
And so we are growing the team
because we are adding and we're
bringing in new tools, and so I think
we'll get into a little bit of what
we're building here in a second.
But we are growing and we're
very specifically growing as
you can imagine, in sort of the
web kind of, side of things.
But also very much on the AI and
kind of agentic side and how we
can, bring all of that together.
And really looking for engineers
who I would say have experience
in either side of that, right?
If you can come in and you're, you've
been in AI and ML and really have
experience there, there's definitely
gonna be a key opportunity I think
probably later this year or soon.
And then also on the other side if
you've got experience with browsers or
if you're alumni from Firefox, actually.
If you are a Firefox alum and you hear
this, please reach out to our team.
Like we-- that's like a target for us
is to bring some, some folks back that
used to work on kind of original Firefox.
Yeah, super small team, but we
bring in really great contractors
like Steve and have them for
long term so we can kick butt.
But small and scrappy in the team size.
Oh, it's so exciting.
Okay, so we're working on
two, arguably three things.
And so I can't give all the details
on all the things, and so I'm gonna
kinda like allude to a few different
things and let folks imagine.
But Mozilla is definitely pushing
for really being a trusted company
in the, in our technical ecosystem.
And so as you think about that details
about Tabstack that we didn't kinda
get into is just really listening to
where we're allowed to be and where
we're not allowed to be, and really
deeply caring about data privacy.
So one thing we didn't say is there's
absolutely zero data retention on
Tabstack, so we'll keep your calls
for 90 days only if we need to support
you in some way, shape, or form.
Those are dumped after 90 days.
It's a marketer's nightmare because
I've gotta take snapshots of results
and understand where things came
in and then make sure that I'm
like, everything is data privacy.
I can't even speak to some of
the results of my marketing work
because I'm like, "I don't know."
The developers don't wanna be
f- they don't wanna be, like,
identified, so we don't know.
And so I think that's really
the beautiful thing there.
And so thinking about kind
of that whole ecosystem and
just awesomeness there, right?
And we're thinking about sort of
agentic systems and on the, the
same line as browsers, right?
We're digging into where does the
browsers go in the new world and
in the agentic world, I should say.
And so just thinking about what
does that infrastructure look like?
What does the browser
experience look like?
What does kinda all
those things look like?
How do we actually evaluate
these types of systems, right?
Going back to Tabstack and our internal
evaluations the whole team, there's a
few folks in Firefox and, a part of some
other organizations that are working on
some really cool evaluation type of stuff.
All those things I can say without the
details, but one of the things will
be launched, I think, later this year,
at least getting into a private beta.
So if that is something that interests
you, definitely reach out to me,
reach out to Steve, or just join
Tabstack or w- join our mailing list
because you'll get notifications
through that channel as well.
So exciting stuff.
I have never seen someone write such
amazing developer content and I cannot say
enough good praises about Steve's content.
For me, yeah, if I am TessaK22 pretty
much across the whole internet.
So if you're on any platform and you
wanna find me that's pretty much it.
I'd say LinkedIn and X are where you
can probably find me most actively.
Yeah, tessacrescel.com
if you're into that.
I don't blog very much, but
there's a few, there's a few blogs
And what about Tab Stack?
So Tabstack's over at tabstack.ai,
so T-A-B-S-T-A-C-K.A-I.
That's pretty easy
Hey, look, it's been amazing
having both of you here.
I appreciate you coming
through and chatting.
Steve MacDougall, Tessa thank you a ton
for joining us for APIs You Won't Hate.
We'll catch you next time.