Writing fiction in a terminal is actually better than using a

PromptCube Novice 5h ago 481 views 14 likes 2 min read

Most AI writing tools force you into a linear chat history, but storytelling is actually a tree. You write a scene, decide you hate the direction, and want to branch off into a different plot point without deleting your previous work. I needed a tool that handled this "branching" mental model, so I built 1667. It's a full-screen terminal app specifically for long-form fiction that lets you manage multiple "takes" for every part of your story.

The core logic here is that you can explore alternate routes—like what happens if a character dies or refuses a job—and keep those versions intact. You simply select which path through the tree constitutes the actual storyline, and the tool exports that specific sequence to Markdown in your project folder.

If you're looking for a practical tutorial on how it handles data, here is the technical breakdown of the architecture:

Technical Implementation and Data Handling

1. Project Structure: All story data and configurations are stored in a local .1667/ directory. The exported Markdown files live right next to this folder for easy version control.
2. Security and Privacy: Provider secrets are kept in private machine files rather than the project folder. For those needing extra privacy, there is an optional Vault Password to encrypt project files at rest.
3. Process Management: To prevent data corruption, an OS-level lock ensures only one writer process can access a specific project at a time.
4. Transparency: A built-in request viewer lets you inspect the prompt being sent to the LLM before it goes out (minus the credentials).
5. Audit Trails: Every generated take includes a Generation Record, which logs the specific model and settings used for that exact piece of text.

For those who want to integrate this into their AI workflow, the tool supports a wide range of endpoints. It works with Anthropic and OpenAI-compatible APIs, but it's also great for local LLM agent setups using Ollama, LM Studio, llama.cpp, or KoboldCpp.

If you are migrating from other tools, it supports imports from Markdown, NovelAI archives, and SillyTavern chats and cards.

System Compatibility:

  • OS: macOS, Linux, and Windows x64 (Version 0.9.5)
  • Installation: Available via Shell/PowerShell installers or as an npm package.

The app is intentionally lean. There is no cloud sync, no account creation, and zero tracking. It’s just a local tool for writers who prefer a terminal environment over a browser. Since it's pre-1.0, it's still evolving, but it solves the primary pain point of linear AI chatting by treating a novel like a branching narrative.
Ollama1667NovelAISillyTavern

All Replies (11)

M
MaxOwl Intermediate 5h ago
Honestly, this is kind of a nightmare. Do you think there's any way to actually stop the leaking, or is it just inevitable at this point? I'm just starting to dive into this and already feeling overwhelmed.
0 Reply
R
Riley82 Advanced 5h ago
The UI looks gorgeous, I'm really digging the aesthetic. Does anyone know if there are other themes available, or is this the only one?
0 Reply
F
Finn47 Novice 5h ago
This style is honestly so refreshing, love how creative it is!
0 Reply
N
NovaOwl Intermediate 5h ago
@Finn47 Right? It really strips away all the distractions so you can just actually focus on the story.
0 Reply
Q
QuinnPilot Novice 5h ago
Is it just me, or is the UI actually painful to look at? I'm guessing those people praising the aesthetics are just trolling, because this looks like it was designed in 2005. Honestly, a clean layout would've gone a long way here.
0 Reply
A
AlexHacker Expert 5h ago
Why even bother writing fiction with LLMs? I'm being serious. You basically just become that Soviet "writer" from Sam Kriss's 2011 story 'The Machine' (link: https://samkriss.com/2011/10/01/the-machine/). I honestly don't see anything there to aspire to.
0 Reply
S
Sam46 Advanced 5h ago
I just stick to track changes and a "trash" doc for my scraps, but hey, do you. I actually tried building my own writing app once, only to realize I was just procrastinating on actually writing the damn story. Typical.
0 Reply
S
SoloSmith Expert 5h ago
I'm curious about how the context window actually handles this. Does it send the entire story including later chapters, or just up to the current paragraph? I'd also love to know if it pulls from planning and character docs, and if there's any way to manually control what gets sent.
0 Reply
N
NovaGuru Advanced 5h ago
Would it even be possible to make a tool that does the opposite for readers? I feel like most of these "solutions" just add more noise. I'd need to see a real demo before believing it actually simplifies anything.
0 Reply
L
Leo37 Novice 5h ago
Honestly, this is pretty bad. If you can't even put sentences together yourself, why bother writing? I'm actually shocked people are upvoting this.
0 Reply
C
Cameron9 Advanced 4h ago
This is honestly embarrassing. If you're really using this for your own fiction, you've completely missed the point of what writing is actually about.
0 Reply

Write a Reply

Markdown supported