zed-industries
Edit Predictions: Edit predictions are unavailable
Summary
Edit predictions are unavailable under the free plan.
Description
Code input in any situation
Actual Behavior:Display "No Prediction" or omit the inline panel entirely. The log shows:
Caused by:error predicting edits.
Status: 403
Body: You must subscribe to a plan in order to use edit predictions. Please visit https://zed.dev/account to start your trial.
Expected Behavior:Get the code predictions provided by the free plan.
Overall, this is similar to the issue mentioned in #30082, but I didn't get a definitive explanation in that issue. I'm not sure if it's due to some glitch preventing the use of the free plan quota, or if the free plan quota has already been exhausted. In the version before the Agentic Editing update, I didn't have any issues using it.
Zed Version and System Specs
Zed: v0.185.12 (Zed) OS: macOS 15.4.0 Memory: 18 GiB Architecture: aarch64
i can confirm this issue also happened with Github Copilot, the prediction and inline completions is not working
Same errors with zed, but I just switched edit_prediction_provider from zed to copilot and inline completion is working as expected.
Zed Preview 0.186.3 (da333757f1442f55aea615a74174090d495c9e86) MacOS 15.4.1 (24E263)
I have found a solution in a discord community. You need to setup this cofig
"features": {
"edit_prediction_provider": "copilot"
},
so copilot is used but is there a general problem with zed completion or is it not free anymore?
Same problem here.
Downloaded the Editor for the first time, logged in and I'm only seeing "No prediction" without using a single edit prediction from the 2000 edit predictions that come from the Free plan.
The other prediction providers (copilot/supermaven) seem to work as expected.
Zed 0.185.13 ( 743f75f743611fbfcd607c23de085092b95fa5c3 )
macOS 15.4.1
I have found a solution in a discord community. You need to setup this cofig
"features": { "edit_prediction_provider": "copilot" },
it works, thank mate! @vadim-su
on trial as for now prediction considered done and therefore count if it only SHOW itself on screen WITHOUT any alt+L pressed to confirm. so while trial time its doesn't really matter, but "free plan" ones will burn out for a hour.
Zed 0.185.13 743f75f743611fbfcd607c23de085092b95fa5c3
Just switched to/trying Zed. Had the same problem, not seeing any predictions. Had to switch to the trial version to get it working. Seems like the free plan has canceled predictions (from Zed/Zeta) all together. Might be smart to mention this somewhere before people give up and return to VSCode/Cursor et al.
Hey folks, sorry for the confusion here.
We historically had a free plan that everyone was automatically on.
This plan has been superseded by the new Zed Free plan which requires an explicit opt-in.
For anyone running into this, if you go to zed.dev/account you will have the option of subscribing to the Zed Free plan, or starting a 14 day Zed Pro trial:
See our pricing page for what's included in each plan ("Zed Free" corresponds to "Personal" on the pricing page).
We are in the process of improving the in-product messaging to help convey the differences clearer and provide a better indication of which plan you are on.
Again, apologies for the confusion around this.
I am using subtle mode and I found my quota burning quickly even if I am not using it at all. I guess it performs predictions when I am typing and the UI is just hiding the prediction results.
Would it be possible to perform predictions only when the option key is held in subtle mode?
I would like to use the edit predictions feature in Copilot. What I am looking for is to auto-correct and skip with tab when I make a function or name change.
really brutal move from Zed here, I never used any of their free chat credits, not interested at all in chat being in the editor, or agents, but edit prediction is handy for boilerplate automation, I used it despite it not being as good as Cursors or even Windsurfs (which has unlimited edit predictions on free tier)
for someone like me to pay 20 a month just for edit prediction is absurd, it would need to be far superior to cursors to even consider it, and its actually nowhere near as good
2000 predictions seem to be used up in no time at all, with most of them wasted/unseen, surely there is enough money in funding with the stupid valuations of these AI editors, and all the vibe coders paying for agents and models to have edit prediction on the free tier, or at least make it 20000 a month
Just for my understanding, what's the purpose of setting "edit_prediction_provider": "copilot"? Do I need to subscribe if I want edit prediction even though I selected copilot as a provider, which I already pay for?
I'm asking this since the above setting does nothing in my case. It's been several weeks since being able to jump around with tab stopped working, and it's not very clear to me what changed and what I have to do to enable it back.
@futurisold do you also have "edit_predictions": { "mode": "eager" } enabled? I have this together with "edit_prediction_provider": "copilot" and predictions/jumps from copilot are working fine, although I see them less often compared to previously using zeta.
cc @ayberkgezer if you haven't figured this out yet.
@futurisold do you also have
"edit_predictions": { "mode": "eager" }enabled? I have this together with"edit_prediction_provider": "copilot"and predictions/jumps from copilot are working fine, although I see them less often compared to previously using zeta. cc @ayberkgezer if you haven't figured this out yet.
Yes, it's enabled. This is my current config:
{
"icon_theme": "Bearded Icon Theme",
"show_edit_predictions": true,
"edit_predictions": {
"mode": "eager",
"enabled_in_text_threads": true
},
"agent": {
"always_allow_tool_actions": true,
"default_profile": "write",
"default_model": {
"provider": "anthropic",
"model": "claude-3-7-sonnet-thinking-latest"
},
"version": "2"
},
"features": {
"edit_prediction_provider": "copilot"
},
"language_models": {
"anthropic": {
"version": "1",
"available_models": [
{
"name": "claude-3-7-sonnet-latest",
"display_name": "Claude 3.7 Thinking",
"max_tokens": 200000,
"mode": {
"type": "thinking",
"budget_tokens": 32368
}
}
]
},
},
"ui_font_size": 15,
"buffer_font_size": 15,
"buffer_font_family": "JetBrains Mono",
"theme": "Zed Legacy: Rosé Pine",
"experimental.theme_overrides": {
"panel.focused_border": "#524f67",
"ignored": "#504c63",
"syntax": {
"comment": {
"font_style": "italic"
},
"comment.doc": {
"font_style": "italic"
},
"constant": {
"color": "#907aa9"
}
}
},
"vim_mode": true,
"vim": {
"use_system_clipboard": "on_yank"
},
"format_on_save": "off",
"enable_language_server": true,
"tab_size": 4,
"relative_line_numbers": true,
"languages": {
"Ruby": {
"language_servers": ["!solargraph"]
},
"Python": {
"show_edit_predictions": true,
"format_on_save": "off",
"enable_language_server": true,
"language_servers": ["pyright"]
}
},
"lsp": {
"pyright": {
"settings": {
"python.analysis": {
"diagnosticMode": "off",
"typeCheckingMode": "off",
"extraPaths": ["/Users/futurisold/devspace/projects/symbolicai"]
},
"python": {
"pythonPath": "/Users/futurisold/miniconda3/envs/symai/bin/python"
}
}
},
"texlab": {
"settings": {
"texlab": {
"build": {
"executable": "xelatex",
"args": ["-synctex=1", "-interaction=nonstopmode", "%f"],
"onSave": true,
"forwardSearchAfter": true
},
"forwardSearch": {
"executable": "zathura",
"args": [
"--synctex-forward",
"%l:1:%f",
"-x",
"zed %%{input}:%%{line}",
"%p"
]
}
}
}
}
},
"use_autoclose": true,
"show_completion_documentation": true,
"scroll_beyond_last_line": "off",
"current_line_highlight": "line",
"hard_tabs": false,
"scrollbar": {
"show": "never"
},
"ensure_final_newline_on_save": true,
"remove_trailing_whitespace_on_save": true,
"inlay_hints": {
"enabled": false
},
"soft_wrap": "editor_width"
}
I hopelessly gave up after I completely wiped everything Zed related on my machine due to fear of something being cached and reinstalled Zed from scratch.
Using it with tab doesn't work for copilot anyway. Eagle and subtl features are for zed. It would be good to use them with copilot. Vscode just added this feature. Zed has been doing this for a long time but since the last membership system came out it doesn't work as well as before.
Using it with tab doesn't work for copilot anyway. Eagle and subtl features are for zed. It would be good to use them with copilot. Vscode just added this feature. Zed has been doing this for a long time but since the last membership system came out it doesn't work as well as before.
@Daste745 do you have a different shortcut from tab?
@Daste745 do you have a different shortcut from tab?
I'm using the default tab shortcut. I think I got the predictions behavior confused. The "Jump to edit" popup sometimes shows even for regular inline suggestions. @ayberkgezer is right that edit suggestions accross multiple lines don't work with "edit_prediction_provider": "copilot".
Hey folks, sorry for the confusion here.
We historically had a free plan that everyone was automatically on.
This plan has been superseded by the new Zed Free plan which requires an explicit opt-in.
For anyone running into this, if you go to zed.dev/account you will have the option of subscribing to the Zed Free plan, or starting a 14 day Zed Pro trial:
See our [pricing page](https://zed.dev/pricing) for what's included in each plan ("Zed Free" corresponds to "Personal" on the pricing page).
We are in the process of improving the in-product messaging to help convey the differences clearer and provide a better indication of which plan you are on.
Again, apologies for the confusion around this.
We've made some additional changes around edit predictions:
- You will now automatically be on the Zed Free plan
- As of Zed v0.187.5 (Stable) and Zed v0.188.1 (Preview), edit predictions only count against your limit when you accept them, so the 2000/mo should go farther
Zed Free Plan: Reality vs Documentation - Feature Limitations & User Experience
Introduction Zed’s marketing materials and documentation promote an editor that supports bringing your own models (LM Studio, Ollama, etc.), full AI‑agent capabilities (refactoring, code fixes, inline assistance), and multiple agent profiles. This promises that users on the free plan using custom/local LLMs should have near‑parity with paid or hosted model use.
What Zed Officially Declares Free plan allows unlimited prompts with your own API keys. Free plan also includes usage quotas (e.g. 2000 accepted edit predictions per month, 50 hosted prompts). The documentation states “Custom Models, Custom Tools” are supported. But the docs also say edit predictions with custom models is not yet supported.
User Experience vs Reality Even with LM Studio and Ollama properly configured, edit predictions / inline code suggestions are non‑functional.
Context menu “AI” options, command palette items like “Explain Selection”, “Refactor”, “Generate Documentation” etc., do not appear.
Profiles (Write, Ask, Minimal) are not showing - only "Disabled". Logs show “provider not configured” or “missing field type” errors when Zed tries to read LM Studio model list. What Is Actually Gated / Limited Edit Predictions with custom models are explicitly not supported yet. Unlimited edit predictions are only available on paid plans.
Some UI features require the AI provider to be fully recognized/parsed, which fails if model metadata isn't exactly what Zed expects.
Profiles UI only becomes usable when provider is validated.
Conclusion & Request
Marketing and documentation should clearly state which AI features are usable with local/custom models on free plan vs which are gated.
Zed should fix issues of missing metadata (or allow models missing certain fields) so custom models are recognized correctly. Restore transparency and user trust.
References
Zed: “The Fastest AI Code Editor” blog post Zed Pricing page GitHub issue #30201 Community feedback (Reddit threads etc.)
