Latest / AI Fire Daily / #135 Robin: Free Claude Code With 1,100+ Models and Billions of Free Tokens? Here’s the Setup

Transcript

#135 Robin: Free Claude Code With 1,100+ Models and Billions of Free Tokens? Here’s the Setup · AI Fire Daily

• 0:00
  Imagine you are finally in the zone. You are
• 0:03
  deep into building that website you've been dreaming
• 0:05
  about. Right. You finally hit that flow state.
• 0:08
  Exactly. You tweak your code. You test a new
• 0:11
  feature. You know, you're running your AI agent
• 0:14
  right there in your terminal. Yeah. And then
• 0:17
  you check your dashboard. Oh, the dreaded dashboard
• 0:20
  check. Right. You have burned through $50 in
• 0:23
  just a few hours. I mean, $200 a month is a lot
• 0:27
  of money. It is a massive barrier. It really
• 0:30
  is. Honestly, it is a steep cost just to use
• 0:33
  an AI coding agent like Claude Code. For everyday
• 0:36
  creators, that financial anxiety is just, it
• 0:40
  is killing innovation. It fundamentally limits
• 0:43
  how people build. When every single token costs
• 0:46
  money, you know, you stop experimenting. You
• 0:48
  stop taking risks with your code because you
• 0:51
  are literally paying for every mistake. Welcome
• 0:54
  to the deep dive. Our mission today is exploring
• 0:57
  a really fascinating step -by -step guide. Yeah,
• 1:00
  this one is huge. We are looking at a tool called
• 1:03
  Free LLM API. It acts as this single gateway
• 1:07
  to combine free credits from multiple AI providers.
• 1:10
  Right. We are talking about unlocking up to 7
• 1:13
  .4 billion free tokens a month. This powers your
• 1:17
  local coding agents completely for free without
• 1:19
  paying for any subscription. Exactly. We are
• 1:22
  breaking down exactly how to set up this system.
• 1:24
  It is basically about combining the free tiers
• 1:27
  of the entire internet into one massive engine.
• 1:32
  I have to make a vulnerable admission here. I
• 1:35
  still wrestle with prompt drift myself. Oh, sure.
• 1:38
  We all do. You know how it is. You tweak a prompt
• 1:41
  to get better code. You're just trying to refine
• 1:43
  a single function. It's one little thing, yeah.
• 1:45
  Right. And you test it out in your terminal.
• 1:47
  The agent loops a few times. And those expensive
• 1:50
  API credits just vanish instantly. It connects
• 1:55
  deeply to the anxiety of paying for every single
• 1:57
  code tweak. That anxiety is a very real part
• 2:01
  of development. It is the friction that stops
• 2:03
  a lot of brilliant ideas dead in their tracks.
• 2:06
  But the sheer scale of what we are unpacking
• 2:09
  today, it changes that cost structure completely.
• 2:13
  Bypassing that financial stress shifts you from
• 2:15
  a scarcity mindset to an abundance mindset. Paying
• 2:19
  per token causes severe developer anxiety. So
• 2:24
  we have to find a reliable way around it. The
• 2:26
  core solution we're exploring today is aggregation.
• 2:28
  Right. Combining those free tiers. Okay. Let's
• 2:32
  unpack this. What does free LLM API actually
• 2:36
  do? Well, think about the current landscape.
• 2:39
  Claude's free plan is very popular, right? But
• 2:41
  it doesn't include access to Claude code or co
• 2:44
  -work. To use those local coding agents you normally
• 2:47
  have to pay. Now, free AI models are absolutely
• 2:50
  everywhere right now. Every tech company is offering
• 2:53
  a free tier as a loss leader. But managing them
• 2:55
  individually is a nightmare. It is a total nightmare.
• 2:58
  Right. Because you would be juggling dozens of
• 3:00
  different API keys. Yeah. You would be rewriting
• 3:03
  your application's connection logic every single
• 3:05
  time one ran out. Which nobody wants to do. Exactly.
• 3:08
  So free LLM API solves this. It brings all those
• 3:11
  free AI models into one single place. You don't
• 3:14
  have to manage separate connections anymore.
• 3:16
  It just centralizes it. Right. It offers access
• 3:18
  to over 1 ,100 different models. And this access
• 3:21
  spans across 76 different AI providers. That
• 3:24
  is a staggering amount of access. And it is not
• 3:27
  just basic models either, is it? Not at all.
• 3:29
  You get access to heavily capable models, things
• 3:31
  like Gemini 3 .8 Flash. You can access QEM 3
• 3:37
  .8 Max. It even includes models like Kimi K3.
• 3:40
  You are getting real production -grade intelligence
• 3:43
  here. It is kind of like stacking Lego blocks
• 3:45
  of data. I love that analogy. Yeah, you take
• 3:48
  a small free quota from one provider. You snap
• 3:51
  it together with another free quota from somewhere
• 3:53
  else. You are essentially building a massive
• 3:56
  free computational reserve for yourself. Whoa.
• 3:59
  Imagine scaling to a billion queries. It's crazy.
• 4:03
  It is actually even bigger than that. You can
• 4:05
  reach up to 7 .4 billion free tokens per month,
• 4:09
  depending on the individual limits of those providers.
• 4:11
  But does this mean you are constantly switching
• 4:13
  contacts manually when one model runs out of
• 4:16
  tokens? That is the best part. You don't have
• 4:19
  to lift a finger. Oh really? Yeah. The single
• 4:21
  endpoint handles the routing automatically for
• 4:23
  you. It monitors your usage behind the scenes,
• 4:26
  and when one provider runs out of credits, it
• 4:29
  instantly switches your prompt to the next available
• 4:31
  one. So the single endpoint acts as an invisible
• 4:35
  traffic cop for your tokens? Precisely. It keeps
• 4:38
  the traffic flowing without you ever noticing
• 4:41
  a jam. Now that we understand what this traffic
• 4:43
  cop does, we need to get the cars on the road.
• 4:46
  Right. We have to actually install the software.
• 4:49
  Exactly. We install it and then hook up the various
• 4:52
  providers to give it that token fuel. And just
• 4:54
  briefly, people can grab this from freellmapi
• 4:57
  .co, right? Yep. You navigate to their GitHub
• 4:59
  releases page. They have standard installers
• 5:02
  like an EXE file for Windows and a DMG for Mac.
• 5:05
  We won't bore you with the standard installation
• 5:06
  clicks. We know you know how to install an app.
• 5:09
  Let's talk about the strategy once it is open,
• 5:11
  because once we launch the app, we have to feed
• 5:13
  it API keys. Right. You go to the keys menu and
• 5:17
  add them. But, you know, you don't need to scour
• 5:19
  the internet for all 76 providers right away.
• 5:21
  Oh, thank goodness. Yeah. The guide highlights
• 5:23
  three specific pillars to get a really strong
• 5:25
  foundation going. The first one is Google AI
• 5:27
  Studio, right? Yes. This is for your heavy lifting.
• 5:31
  You sign in and create a cloud project. That
• 5:33
  gives you access to several free Gemini models.
• 5:35
  OK. So you just click to create your API key
• 5:38
  and drop it into the app. Simple as that. The
• 5:41
  second provider on the list is NVIDIA NIM. NVIDIA
• 5:44
  NIM is fantastic because it offers around 17
• 5:47
  available models right out of the gate. 17. Yeah.
• 5:50
  And these are often specialized, highly optimized
• 5:53
  models. You just visit their build site, create
• 5:56
  an account, and generate the key from your profile.
• 5:58
  And the app has a way to test that, right? Right.
• 6:00
  The free LLM API app actually has a convenient
• 6:04
  check button to test this connection immediately.
• 6:07
  The third provider is called OpenRouter. I've
• 6:09
  always found OpenRouter to be a really fascinating
• 6:11
  platform on its own. OpenRouter is a fantastic
• 6:14
  resource for developers. It acts as an aggregator
• 6:17
  itself, offering access to the open source community.
• 6:20
  Oh, cool. It provides around eight different
• 6:22
  free models right now. This includes powerful
• 6:24
  open weights like Gemma 4 and Nematron 3 Nano.
• 6:28
  Nice. You create a key there, and you can even
• 6:31
  set spending limits to ensure you strictly stay
• 6:33
  on their free tier. So if we look at the math
• 6:36
  here. Together, just these three providers alone
• 6:39
  can reach over a hundred million tokens. It is
• 6:41
  an incredible amount of coding power. You could
• 6:44
  probably rewrite a massive code base with that,
• 6:46
  but there is a crucial tip we must mention here.
• 6:49
  You must keep your API keys entirely private.
• 6:52
  Absolutely. Never commit them to a public GitHub
• 6:55
  repository. Right. And always check each provider's
• 6:58
  specific free limits because they do change.
• 7:00
  Yeah. Security is paramount. Yeah. Treat those
• 7:03
  keys like passwords to your bank account. I have
• 7:06
  to play devil's advocate for a moment. Bring
• 7:08
  it on. I know we just talked about the top three,
• 7:11
  but the platform supports 76 different providers.
• 7:15
  Managing API keys for 76 different providers
• 7:17
  sounds incredibly tedious. It does sound like
• 7:20
  a lot. Is the juice really worth the squeeze
• 7:21
  here? Are developers actually going to spend
• 7:23
  their weekend hunting down 76 different API keys?
• 7:28
  What is fascinating here is the massive return
• 7:30
  on your time. You don't need to connect all 76
• 7:33
  providers to get value out of this. OK, that
• 7:36
  is a relief. A one -time setup of just those
• 7:38
  three providers unlocks a massive recurring benefit.
• 7:42
  You spend maybe 15 minutes setting up Google,
• 7:45
  NVIDIA, and OpenRouter. Right. And in return,
• 7:47
  you get 100 million tokens that refresh constantly.
• 7:51
  That is months of free coding power for a few
• 7:53
  minutes of administrative work. How can we be
• 7:56
  absolutely sure the API connections are live
• 7:58
  before we start trying to code with them? Oh,
• 8:01
  it is very simple to verify. Because I would
• 8:04
  hate to get deep into a coding session only to
• 8:06
  realize the gateway is broken. Exactly. You use
• 8:08
  the Models tab inside the free LLM API app to
• 8:11
  check your available quotas. It will show you
• 8:13
  exactly what is loaded. OK. Then you open the
• 8:16
  Playground tab right there in the app. You send
• 8:19
  a very simple message to test the routing. Like
• 8:21
  what? You just ask, hi, which model are you?
• 8:24
  If you get a response back, say from Gemini or
• 8:27
  Nematron, the connection is live and working.
• 8:29
  Just ping the Playground first to verify the
• 8:31
  models are actually talking back. Exactly. It
• 8:33
  is a quick sanity check. With the tokens successfully
• 8:37
  flowing into the app, we have to shift to the
• 8:39
  final bridge. This is where it gets really exciting.
• 8:42
  It really is. We need to pipe this free intelligence
• 8:44
  directly into Claude's ecosystem. That is how
• 8:47
  you actually start building your projects in
• 8:49
  your own terminal. Right, because right now the
• 8:51
  free LLM API is just sitting there holding the
• 8:53
  tokens. Yeah, doing that thing. We need to tell
• 8:56
  Claude code how to find it. We need to get the
• 8:58
  unified API key first, and we also need the base
• 9:01
  URL from the app. Yes. You go to the keys menu
• 9:04
  to find those two crucial pieces of information.
• 9:06
  OK. The unified API key is essentially your master
• 9:10
  skeleton key. It connects to all your added providers
• 9:12
  simultaneously. And the base URL is the endpoint
• 9:16
  your coding agents will use to communicate with
• 9:18
  that skeleton key. Let me stop you right there
• 9:20
  for a second. Sure. We hear the term endpoint
• 9:23
  a lot. in software development. How would you
• 9:27
  define that simply? A digital doorway where two
• 9:30
  software programs talk to each other. That makes
• 9:32
  perfect sense. So we take that digital doorway,
• 9:35
  that base URL. We open the Anthropic tab in the
• 9:38
  free LLM API app. Right. And we see configuration
• 9:42
  slots there for Opus, Sonnet, and default models.
• 9:45
  The guide recommends setting all of those slots
• 9:47
  to auto. That is a very important configuration
• 9:49
  detail. Why is that? You are basically telling
• 9:52
  the app how to map Claude's expected models to
• 9:55
  your free ones. Next, we move to the Claude code
• 9:57
  setup itself. We are actually in the terminal
• 10:00
  now, right? Yes. You find Claude code in the
• 10:03
  Agents tab of our free app. You copy the automatic
• 10:06
  setup command provided right there, but there
• 10:09
  is a very crucial step before running it in your
• 10:12
  terminal. Yeah, this part is vital. You have
• 10:13
  to check your Claude code configuration file.
• 10:16
  It is usually located in your user directory
• 10:19
  at dot Claude slash settings slash JSON. Yes.
• 10:25
  Why do we need to touch this specific file? Right.
• 10:28
  And you must back up that specific file first.
• 10:32
  Claude code is incredibly stubborn. Oh, I know.
• 10:35
  If it sees old API settings in your JSON file,
• 10:37
  it will try to route through Anthropics paid
• 10:40
  servers by default. So it will just ignore our
• 10:43
  free setup? Exactly. You have to remove any old
• 10:47
  conflicting API settings. We have to wipe the
• 10:49
  slate clean so it accepts our new free gateway.
• 10:52
  OK, got it. Only then should you run the new
• 10:54
  setup command. OK, so we've cleared out the old
• 10:58
  page settings. You open a terminal and paste
• 11:01
  that setup command we copied earlier. Yep. You
• 11:04
  type Y to confirm when it prompts you. Then you
• 11:08
  paste your unified API key, that skeleton key,
• 11:10
  into the terminal. Right. Oh. And just a heads
• 11:13
  up for anyone newer to the terminal, the key
• 11:16
  actually won't appear as you type it. It is hidden
• 11:19
  for security. Once the setup finishes, you simply
• 11:22
  run the command Claude. The tool will boot up.
• 11:25
  It is that easy. If it asks you to trust the
• 11:28
  current folder, Confirm it, but obviously only
• 11:30
  if it is a folder you actually trust. Right,
• 11:33
  because you are giving an AI access to read and
• 11:35
  write files there. Exactly. Then you can test
• 11:39
  it with a real prompt, something you would actually
• 11:41
  use in your day -to -day work. Like what? You
• 11:43
  ask it to build a modern landing page, maybe
• 11:46
  something for a personal portfolio website. Just
• 11:50
  watch it generate all the HTML and CSS using
• 11:52
  your free tokens. It is incredibly satisfying
• 11:55
  to watch. You can also connect this entire system
• 11:58
  to Cloud Desktop if you prefer a graphical interface
• 12:00
  over the terminal. Oh, that is good for people
• 12:02
  who don't want to live in the command line. Exactly.
• 12:05
  You go to the menu in Cloud Desktop, then Developer,
• 12:08
  then click Enable Developer Mode. The application
• 12:11
  will restart itself after you do this to apply
• 12:13
  the changes. Then you click Configure Third Party
• 12:16
  Inference. Right. This is where we tell the desktop
• 12:18
  app to look at our digital doorway. You set the
• 12:21
  connection type strictly to gateway. The credential
• 12:24
  kind needs to be set to a static API key. You
• 12:27
  input your base URL and your unified API key
• 12:30
  into the field. And the authentication scheme
• 12:33
  must be set to bearer. Yes. Bearer authentication.
• 12:36
  It's basically just a standard digital handshake
• 12:39
  telling Claude you have the right access token,
• 12:40
  right? Exactly. It just proves you are allowed
• 12:43
  to use that endpoint. Click apply changes, then
• 12:46
  save and restart. And now your desktop app is
• 12:49
  wired into the free grid. Why is setting the
• 12:52
  model slots to auto so critical during this setup?
• 12:55
  We mentioned it earlier, but what is the actual
• 12:58
  mechanism happening there? Because if a specific
• 13:00
  provider's quota hits zero, auto kicks in automatically.
• 13:04
  Let's say you burn through all your Google Gemini
• 13:06
  credits for the day. Okay. Auto allows the free
• 13:09
  LLM API to seamlessly switch without throwing
• 13:12
  an error in your terminal. It immediately routes
• 13:15
  you to the next available free model in your
• 13:16
  stack, like OpenRouter. That is seamless. For
• 13:19
  example, if Claude expects Opus and Opus is mapped
• 13:22
  to Gemini 3 .8 Flash, requests using that slot
• 13:25
  will be routed directly to Gemini. If Gemini
• 13:28
  is out of juice, the invisible traffic cop points
• 13:31
  it down the next open road. So Auto Mode guarantees
• 13:34
  your coding agent never hits a dead end when
• 13:36
  credits run dry. That is exactly right. It keeps
• 13:40
  your workflow completely uninterrupted. You just
• 13:42
  stay in the zone. So jumping back in, we have
• 13:50
  successfully bypassed the $200 subscription.
• 13:53
  We have. The agent is running. The tokens are
• 13:56
  free. Let's pull back and look at the macro picture
• 13:59
  now. Okay. We are using a bit of a Frankenstein
• 14:03
  approach here, right? We have stitched together
• 14:05
  brains from Google, NVIDIA, and OpenRouter to
• 14:08
  mimic Clot. Yeah, that is a fair way to describe
• 14:11
  it. What are the actual trade -offs of this coding
• 14:14
  setup? There are definitely some clear trade
• 14:16
  -offs to consider. It is not a perfect magic
• 14:19
  trick. Free models are simply not as smart as
• 14:22
  Claude's top -tier premium models, like Claude
• 14:25
  3 .5 Sonnet or Opus. The premium models handle
• 14:28
  complex reasoning, deep architectural logic,
• 14:31
  and massive context windows much better. Exactly.
• 14:34
  Provider usage limits can and will change over
• 14:37
  time, too. You are relying on the generosity
• 14:39
  of these free tiers, which exist entirely as
• 14:42
  marketing loss leaders. That is very true. Their
• 14:45
  usage limits are constantly shifting. What is
• 14:48
  free today might be paywalled tomorrow. Right.
• 14:51
  But for everyday coding, they are highly effective.
• 14:54
  If you are building standard websites, writing
• 14:56
  scripts, or doing data transformation, they work
• 14:58
  perfectly. Yeah. If you are just testing new
• 15:01
  ideas and doing rapid prototyping, they are incredibly
• 15:04
  useful. So what does this all mean? It means
• 15:08
  you don't always need top -tier intelligence
• 15:11
  for boilerplate tasks? No, you really don't.
• 15:14
  A lot of everyday coding is just standard structure.
• 15:16
  It is repetitive. It is writing a basic navigation
• 15:19
  bar or a standard database query. You don't need
• 15:22
  a super genius AI to write a navigation bar.
• 15:26
  You really don't need the smartest AI model in
• 15:28
  the world. Well, unless you're trying to hack
• 15:30
  NASA with HTML. Exactly. For the vast majority
• 15:34
  of us, the free models are more than capable
• 15:36
  for everyday work. It is absolutely worth trying
• 15:39
  this setup out for yourself. It democratizes
• 15:42
  access to these powerful coding agents. It gives
• 15:45
  you a single API connection. You use free AI
• 15:48
  models right inside your preferred coding tools.
• 15:50
  And you do it all without paying for those specific
• 15:53
  model requests. It is a brilliant workaround.
• 15:56
  It really is. Thank you for joining us on this
• 15:58
  deep dive. We have explored a truly fascinating
• 16:01
  alternative today for developers looking to build
• 16:04
  without breaking the bank. I want to leave you
• 16:06
  with a final provocative thought. If we step
• 16:08
  back, We are entering an era where basic AI reasoning
• 16:12
  is effectively free. Right. General coding ability
• 16:15
  is rapidly being commoditized across dozens of
• 16:18
  different providers. If a stitched together network
• 16:21
  of free models can build standard software, how
• 16:23
  will premium models justify their steep subscriptions
• 16:26
  in the future? That is a great question. What
• 16:28
  happens to the economics of software development
• 16:30
  when the free baseline becomes smart enough to
• 16:34
  build almost anything? That is the big question
• 16:38
  we are all facing right now. The landscape of
• 16:41
  software development is shifting rapidly beneath
• 16:43
  us. It sure is. The floor of what is possible
• 16:46
  for free is rising every single day. Keep exploring
• 16:51
  these new tools on your own. Keep building. We
• 16:53
  will see you next time.