Fig Terminal Auto-complete Review

Fig Terminal Auto-complete Review

My entire career there has been a wide gap between CLI applications and GUI applications. GUIs are easier to use at first, they have options to provide more direct feedback to users, but they also can have odd or confusing behavior. Endlessly spinning wheels, error messages that don't mean anything to the user (or don't output any text) and changing UIs between versions has long meant that while GUIs might be easier to use at first they're a less stable target.

CLIs tend to be more stable, are easy to automate against and don't tend to experience as much mysterious behavior. This is why sys admins back in the day would always try to do as much as they could through PowerShell or bash even if a GUI was present. It was just a much better experience....once you got over the initial learning curve. Discovering the options, arguments and what they did was often a time-consuming task that exceeded the technical skill of a casual user.

At first Fig presents itself as just an auto-complete tool and, don't get me wrong, if that is all you are looking for, it does a great job. However there is clearly a much larger ambition with this product to expand the idea of terminal auto-completion into making CLIs much more approachable, baking documentation into the terminal and then allowing you to share that among your entire team for public or private tools. While at present the idea is just starting to become fleshed out, there is a lot of promise here.

So while Fig is a useful tool today, the promise of where Fig could go is really exciting to me. I'm going to focus mostly on what the product is now, but felt I needed to mention the obvious hooks they are building into the application and the overall framework they are constructing.

Disclaimer: I haven't been asked to write this, nor paid to do so. I have zero affiliation with Fig and I'm not important enough to "move the needle" on them.


Fig is a Node server/Rust client local application that runs on your local machine and inserts autocompletes for most of the commonly used CLIs and programming languages. It's basically an IDE autocomplete but across a wide range of standard tooling. It has well-written and clearly defined privacy rules, a robust selection of settings for personalization and in my testing seemed....remarkably feature complete. I have no idea how they intend to make money with this product, but give it a shot before they run out of VC money and end up charging you per autocomplete or something. Grab your download here.

What is auto-completion?

When we say "auto-completion", what we typically mean is one of two things for programmers. The first is the interface of an IDE like the one I use somewhat frequently, PyCharm. You start typing the name of something, hit ⌃ Space and bam you'll get a list of methods as shown below:

Now obviously IDEs do quite a bit more than just give you a list of methods, some apply machine learning to your project to improve the quality of the recommendations as well as some pretty impressive framework-specific recommendations. This is often provided as a compelling reason to adopt IDE-specific workflows vs more traditional text editors. The point being the IDE, being aware of the syntax of that language, attemps to provide common situation-specific completions, be it file paths, methods, dictionary keys or whatever.

The other common place we see autocomplete is in the shell, be it bash or something else. For a long time the standard way in Bash to enable the famous "tab complete" functionality was through the use of the complete builtin. Basically you pass the name of the command, if the command has a compspec then a list of possible completions for that word are provided. If there is no default compspec, Bash will attempt to expand any aliases on the command word and then build off of that to find shortcuts.

You can also define your own in a simple bash script you source. The process is simple and reliable:

#/usr/bin/env bash
complete -W "word1 word2 word3" program_these_words_will_be_used_with

The you just run source ./name-of-file.bash and you will be able to use tab complete with the custom program program_these_words_will_be_used_with. You can get an extremely well-writte write up of how all this works here. Now some programs support this tab-complete behavior, some do not but typically you need to know the first letter of the argument to use it effectively.

Where does Fig fit in?

If your IDE or plugin for Vim/Emacs provides the auto-completion for your editor, what exactly does Fig do? Basically it is bringing the IDE experience of completion to the terminal emulator. Instead of getting just a basic tab-complete you get something that looks like the image below:

Not only do you get a really nice breakdown of all the arguments but you get a quick explanation of what it does along with what to pass to the command. In short, Fig is trying to replace my normal workflow of: read the man page, attempt the command, open the man page back up, second attempt, etc. Instead you get this really fast autocomplete that does feel a lot like using an IDE.

Plus you don't need to switch your terminal emulator. It worked out of the box with iTerm 2 and the setup was pretty straight forward. We'll get into that next, but if you still aren't sure where the value is, check out this screenshot of the aws cli, a notoriously dense CLI interface:

It's just a faster way to work in the terminal.

Installation and Customization

Installing Fig is pretty easy. I'm typically pretty paranoid about installing stuff that is running in my terminal, but I have to give a lot of credit to the Fig team. Their data usage explanation in the setup of the app is maybe the best I've ever seen.

All of this is explained in the installer. A++ work

There are some parts of the setup I didn't love. I didn't like that the download didn't have a version number. Instead I'm downloading "latest".

I'd love a version number

In terms of running the app, it's a menu bar app on the Mac. However the settings menu is a real delight. You have some interesting concepts I haven't seen before in a tool like this, like a "Plugin Store".

Tons of respect to the Fig team for the ease of turning on and off things like "updates" and telemetry. It's all laid out there, not hidden in any way and have plain English explanations of what they're doing. I especially like being able to turn off AutoUpdate. Sometimes these tools get so many updates that seeing the prompt for a new update every time I open the terminal starts to make me go insane. I'm looking at you Oh My Zsh.

You can also selectively enable or disable Fig for different terminals. I like this, especially if you run into an issue and need to troubleshoot where the problem is. Plus Fig provides debug logs to you as a relatively top-level option, not deeply nested.

This is how it looks by default

I appreciate how easy this was to get to. I haven't actually encountered any problems with Fig that required debugging, but especially for a tool that is this "in the critical path" it's nice to know I don't have to go hunting around endlessly looking for "is there something wrong with Fig or did I forget how this application works".


So the auto-completes you are loading live here. They're open-source which is nice, but you'll notice that Fig itself doesn't appear to be an open-source app. Their documentation also provides tons of great information on how to write your own completes, which seems to be what the tool is geared towards. You can check out that tutorial here. Big shoutout to the Fig documentation folks, this is really well-written and easy to follow.

Fig also supports you writing completes for internal tooling or scripts and sharing them with a team. I assume this is the plan to generate revenue and, if so, I hope it works out. I didn't get the chance to try this, in part because it wasn't clear to me how I join a team or if the team concept is live. But you can read their steps on the documentation site here to get a sense of how its supposed to work in the future.

I definitely could see the utility of syncing this sort of stuff across a team, especially with documentation baked in. It is just so much easier to write a script or CLI to interact with an internal API or database as opposed to attempting to ship a GUI, even with some of the simplistic tooling that exists today. I'll be keeping an eye on Fig to see when and if they ever expand on this. (If someone from Fig wants to invite me to a team or something, I'd love to see it.)


Like I stated before, Fig doesn't appear to be open-source, although let me know if I missed a repo somewhere. Therefore in order to learn more about how it works we need to dig around the app bundle, but other than learning ReactJS and Swift are involved I didn't get a lot of data there. I decided to poke around the running process a bit.

Going through the debug logs it appears the client is a Rust app:

2022-05-05T07:44:05.388194Z DEBUG send_operation{operation="InitiateAuth" service="cognitoidentityprovider"}: rustls::client::hs: 584: Using ciphersuite Tls12(Tls12CipherSuite { suite: TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256, bulk: Aes128Gcm })
2022-05-05T07:44:05.388305Z DEBUG send_operation{operation="InitiateAuth" service="cognitoidentityprovider"}: rustls::client::tls12::server_hello: 82: Server supports tickets
2022-05-05T07:44:05.388474Z DEBUG send_operation{operation="InitiateAuth" service="cognitoidentityprovider"}: rustls::client::tls12: 410: ECDHE curve is ECParameters { curve_type: NamedCurve, named_group: secp256r1 }
2022-05-05T07:44:05.388572Z  WARN send_operation{operation="InitiateAuth" service="cognitoidentityprovider"}: rustls::check: 66: Received a ServerHelloDone handshake message while expecting [CertificateRequest]
2022-05-05T07:44:05.388648Z DEBUG send_operation{operation="InitiateAuth" service="cognitoidentityprovider"}: rustls::client::tls12: 694: Server DNS name is DnsName(DnsName(DnsName("")))
2022-05-05T07:44:05.492802Z DEBUG send_operation{operation="InitiateAuth" service="cognitoidentityprovider"}: rustls::client::tls12: 1005: Session saved

I suspected from the configs being Nodejs that the local server the CLI is interacting with is also Node, but to be fair I'm not sure. If anybody has more detail or documentation I would love to see it.

Overall as an app it seems to be a good citizen. Memory and CPU usage were good, nothing exciting to report there. I couldn't detect any additional latency with my terminal work, often a concern when introducing any sort of auto-complete. I tried to pull timings but due to how you interact with the software (with the interactive drop-downs) I wasn't able to do so. But in general it seemed quite good.

The only criticism I have of the Fig app is the Settings menu isn't following any sort of standard MacOS design template. First, you don't usually call it "Settings" in MacOS, it's "Preferences". Also arranging the options on the side is not typical, you usually make a horizontal line of options. That isn't a hard and fast rule though. One thing I did notice that was odd was it seemed "Settings" is a web app, or loading some sort of resource. If you close it and reopen it, you can see it loading in the options. Even if you just click around there was considerable lag in switching between the tabs at time.

This is on a 16-inch MacBook Pro with 32 GB of RAM, so I would expect this experience on other hardware as well. But you don't need to spend a lot of time in Settings, so it isn't that big of a deal.


Fig is good as a tool. I have zero complaints as to how it works or its performance and I definitely see the utility in having a new and better way to write auto-completes. I worry there is such a huge gap between what it does now (serve as a way to allow for excellent auto-completes for CLI apps) and a business model that would make money, but presumably they have some sort of vision for how that could work. If Fig becomes popular enough to allow developers inside of a company to ship these kinds of detailed auto-completes to less technical users, it could really be a game changer. However if the business model is just to ship these custom configs for internal apps to technical users, I'm not convinced there is enough utility here.

But for you on a personal laptop? There doesn't seem to be any downside. Good privacy policy, excellent performance and overall a low-stress setup and daily usage is impossible to beat. This is just pretty good software that works as advertised.