Tidewave Web: in-browser coding agent for Phoenix (and LiveView)

Chat & Discussions Chat / Discussions
,
1

Hi everyone!

The team at Dashbit have been working on a coding agent that is deeply integrated with Phoenix (and also Rails) and understands your web application, from the database to the UI. We have just launched it and we hope you will enjoy it. Here is the announcement: Tidewave Web: in-browser coding agent for Rails and Phoenix - Tidewave

33 Likes
Tidewave has just been announced by José Valim
2

I tested it and it’s fantastic, I hope to use it more in the future, but I confess that paying for another paid service in dollars while earning in reais makes me take a deep breath, but it’s very much worth it. :smiling_face:

3 Likes
3

Congratulations on the launch!

Since I assume some (many?) of us are already using monthly subscriptions like Claude Code Pro/Max, I’m wondering how hard would it be to use Tidewave Web with Claude Code SDK?

I will be happy to pay Tidewave subscription, but paying for all the extra Anthropic tokens while I already have CC would be tough.

3 Likes
4

Same here. However once openrouter support is in place for models like kimi2 or qwen3 I am ready to jump on.

I know opencode supports using cc login, but I suspect that is reverse engineered and outside ToS :confused:

1 Like
5

Very interesting. we have @chrismccord launching https://phoenix.new a few weeks back and now @josevalim launching https://tidewave.ai. I am reasonably sure, both approaches use GitHub - tidewave-ai/tidewave_phoenix: Tidewave for Phoenix under the hood. First, these two dabbling their hands at this means two things.

  1. Agentic coding is going to be a real thing.
  2. Elixir is somehow more suited to this than many other languages.
    I have been contemplating for sometime now what makes a good agent. What is the best way to make the agentic coding optimal. When we say optimal agentic coding optimal, we again mean these things:
  3. Generate correct code.
  4. Use optimal number of tokens.
  5. Generate code by keeping the current project in mind - not some generic patterns.
  6. Optimal edits.
    For attaining these things essentially improvements can come in two directions:
  7. How to prepare better context - so that agent can send the optimal prompt to LLM.
  8. How to reduce the plies/turns between agent and human interaction cycles.
    Now, step 1 - providing better context is done by the open sourced GitHub - tidewave-ai/tidewave_phoenix: Tidewave for Phoenix
    For step 2 - providing a browser to be aware of what is happening - that is where both tidewave and phoenix.new are coming up with their own takes. It is more like some playwright integration into the agent - for reducing turns between agent and human interaction cycles.
    I believe efforts in these two areas will keep on easing the developer pain - and are mandatory for vibe coding to morph into real coding.
    The inbuilt AST of Elixir, very stable patterns in coding, the functional nature combined together place Elixir, Phoenix and LiveView with a distinct advantage in building real time applications in the Agentic Era.
    Exciting times ahead.
4 Likes
6

If you are paying for Pro/Max, please reach out to Claude and let them know you would like to use your Claude Pro/Max subscription with other coding services like Tidewave. Unfortunately we can’t use the Claude Pro/Max APIs according to their ToS.

5 Likes
8

After upgrading to 0.4 the /tidewave/mcp endpoint no longer works?

9

v0.4 updated the protocol to streaming http. So make sure your editor supports it and you update the MCP-proxy, if you’re using one.

10

For me this means:

claude mcp remove tidewave
claude mcp add --transport http tidewave http://127.0.0.1:4000/tidewave/mcp

Thanks @josevalim

8 Likes
11

I’m already subscribed, and my experience so far has been amazing!! Adding new features has never been this easy—thank you so much for the work you’ve put into this!!! My congratulations!! :tada:

3 Likes
12

I am actually blown away by this. I have been using Elixir since early 2017 and since around phoenix 1.2. I retired around 2018 and since then haven’t worked in a professional setting or with other developers so keeping up with things became more difficult than when I had a job as a Rails dev. The last few years I have been making phoenix apps by myself for people mostly through word of mouth, but once vibe coding really became a thing I kind of ignored it because I was already in my groove. This is my first taste of ai mixed in with the development beyond asking Claude quesitons, and I think I got 3 weeks worth of work done on an app last night alone. This thing is coo!

8 Likes
13

I’ll definitely give it a try too!! Do you use github copilot or Claude Api?

14

I just subscribed to the paid Tidewave pro plan.

I mainly use aider but there is an accessibility gap there when it comes to changing UI.
When I tried Tidewave out it filled in that gap for me so I completed some tasks that have been open for a while now.

I already pay for GitHub Copilot monthly so the fact that I can use that with Tidewave made it just the right tool for me and the additional cost is actually going to be probably less money then using Claude Code CLI with our Claude API key which I haven’t even started since I started using Tidewave.

So far very good.

3 Likes
15

I had copilot already and it’s working great with gpt-5 mini. I am really curious about how Claude compares though and have considered dropping $5 to test it out.

16

I already have a Claude Code Pro plan and I’m really happy with it for adding new features to my Phoniex apps. Now I’m really curios about Tidewave but paying for using Claude API too is a little uncomfortable…

1 Like
17

@josevalim I’m digging tidewave web so far, and I’m curious about a public changelog, as this is likely(?) different from the standard tidewave_phoenix changelog, and there’s nothing mentioned on the https://tidewave.ai/ site. I think this would be a positive thing to add, so that anyone evaluating tidewave web can see the progress transparently.

Additionally, what’s the best way to request and discuss features specifically for the tidewave web side of things?

18

If you have copilot you can enable Claude in your copilot settings. :slight_smile:

We will publish updates on the website blog: News - Tidewave . First changelog will be next week.

We will also open up a public roadmap. But the best way to request a feature is to discuss it here or in our discord. We are constantly evaluating user feedback and prioritizing accordingly.

2 Likes
19

Maybe the Github docs should specify the minimum Elixir version (1.18 i think). Was running on 1.15 and file read was not working.

Otherwise browser integration seems incredible.

Cheers

20

Feature Request: more granular agent permissions

At the moment, tidewave web either requires manually approving all actions taken by the agent, or bypassing checks altogether via the Setting’s Auto-run mode toggle. I imagine more granular permissions are already on the roadmap, but I wanted to mention the need explicitly so that folks can choose to upvote it to influence its priority, as approving is tedious for innocuous commands, and the ‘Auto-run mode’ is a no-can-do for many folks who are just evaluating the tool or have yet to containerize its execution.

5 Likes
21

Earlier Elixir versions should work. Which error in particular were you seeing?