This feature is available only on the Enterprise plan and is currently in beta. To test drive this feature, please reach out to us directly.
Custom AI profiles improve AI output by retrieving examples from your past work and applying them to new strings. Depending on the setup, they can use:
You assign AI profiles to one or more Lokalise projects. Once assigned, AI translations in those projects will use the chosen context to stay closer to your tone of voice and past translations.
For the best results, combine AI profiles with glossaries, style guides, and text descriptions. These features work together—they don't replace each other.
What AI profiles are
An AI profile tells Lokalise what extra context to send to the AI during translation.
By default, the AI already uses these sources of context automatically (not set in profiles):
Style guides
Task descriptions (when assigning AI translation tasks)
Glossaries and text descriptions
However, the AI does not use translation memory (TM) entries or any existing translations in your project. If no custom profile is set for a project or language pair, Lokalise uses the default AI setup without this context.
When creating an AI profile, you choose one source language and one or more target languages. Based on that setup, Lokalise can:
Pull translation memory (TM) entries for the selected language pairs. These entries come from the translation memory storage currently assigned to the project.
Pull existing translations already made in your project for the chosen languages. Any previously translated keys in the team projects can be used as context.
These entries are then passed to the AI so it can follow your past translations more closely and stay consistent with your style.
When to use AI profiles
AI profiles are most useful when you want translations that stick to your style, product wording, or industry tone. They work best for:
UI strings and repetitive patterns
Content with small variations (numbers, units, placeholders)
Projects where consistency with past translations is critical
Example: Similar strings, consistent results
Suppose your translation memory includes these entries:
"Turn ahead in 400 meters" — already translated
"Turn ahead in 500 meters" — already translated
Now you add: "Turn ahead in 450 meters."
Even though it’s new, the AI profile provides the 400/500-meter examples to the model. The output will follow your established phrasing, only changing the number.
Choosing the right AI profile type
Use TM-based profiles when:
Your translation memory is clean and consistent — no duplicates, conflicting entries, or low-quality translations imported from unverified sources.
You want the AI to follow established phrasing across multiple projects or products.
You rely heavily on repetitive content like UI strings, templates, or structured data.
Best for teams with clean, centralized TM data.
Use existing translations-based profiles when:
Your translation memory is incomplete or outdated, but you have high-quality translations already in projects.
You need project-specific tone, for example, marketing or region-focused content.
You prefer to handpick sample keys via tags or reviewed status to keep control over the examples.
Best for teams with strong in-project translation quality and distinct brand voice.
Combine wisely
You can maintain separate AI profiles for different use cases: for example, one TM-based profile for product UI, and one existing-translations-based profile for marketing copy.
Clean, purpose-specific data always produces better results than trying to cover everything with one large dataset.
How AI profiles work under the hood
When you run AI translations or accept AI suggestions, Lokalise fetches similar entries from your translation memory (TM) or existing translations (depending on your profile setup) and sends them to the model. The model then produces a translation that follows those examples.
This process is called retrieval-augmented generation (RAG). Instead of relying only on the model’s general knowledge, AI retrieves context from your own data (past translations, glossaries, style guides, and key descriptions) to generate results that match your brand and tone.
If different sources conflict, the model chooses how to balance them. In practice, TM entries and existing translations usually have the greatest impact on results.
AI profiles with Translation memory
Before you begin: Check assigned translation memory storages
Custom AI profiles can use translation memory (TM) entries as extra context. Before setting up a TM-based profile, make sure the project has the correct TM storage assigned.
To achieve that, open your Lokalise project and proceed to More > Settings:
Scroll to the Translation memory priority section.
Here you’ll see all TM storages available in your team.
If a storage is checked, it’s assigned to this project.
Assigned storages are the ones the AI profile will use.
By default, you’ll see a single storage called Lokalise Translation Memory, but you can create more in Team settings (see Translation memory documentation for details).
What if I have multiple translation memories?
That’s fine. If a project has more than one TM attached:
The AI profile will use all of them for context.
Their priority depends on the order in your project settings.
You can reorder storages by dragging and dropping them.
Don’t forget to save changes after reordering.
What if I have multiple projects that should use different TMs for context?
You do not need to create separate AI profiles for each project. One AI profile can cover multiple projects, even if each project uses a different translation memory.
Here’s how it works:
Each project in Lokalise has its own TM assignments (set in More > Settings > Translation memory priority as explained in the section above).
When you assign a TM to a project, the AI profile will automatically pull entries from that TM when used on that project.
This means the same AI profile can behave differently depending on which project it’s applied to.
Example setup:
You have two projects: Software and Marketing.
You also have two storages: Software TM and Marketing TM.
In the Software project settings, assign Software TM.
In the Marketing project settings, assign Marketing TM.
Create a single AI profile for Software and Marketing projects. While both projects share the same AI profile, the AI will use the correct TM for each project automatically.
Rollout tips
Start small. Enable AI profiles for a few language pairs or one pilot project first. Review the results, then expand to more projects.
Keep glossaries and style guides updated. These work alongside AI profiles and improve output quality.
Clean up your TM. Remove duplicates, outdated terms, and inconsistent phrasing before relying on it as context.
How to enable TM-based AI profiles
Click on your avatar in the side menu and proceed to Team settings > Custom AI profiles.
You’ll see a base profile (non-customizable) that's used by default. The base profile includes style guides, glossaries, and descriptions. It does not include TM entries or existing translations.
On the same page, you can activate a TM-based profile.
You'll see the following dialog:
Choose the source language and the target languages you want this profile to cover.
Start with a few high-traffic target languages, evaluate results, then expand.
Optionally, add a TM date filter if you only want examples from a certain date onward.
Good examples lead to good output. If your translation memory has outdated or inconsistent entries, quality may drop. If you know your TM has improved recently, use the date filter to limit examples to your cleaner period (see below).
Hit Next to proceed to the next step:
Select the projects where you want this profile active.
Click Activate.
The profile's status will show as Active, along with the list of projects using it:
When TM entries may not be a good fit
Translation memory is a powerful data source but only when it’s clean and consistent. AI learns directly from your examples, so low-quality or mixed-content segments can negatively affect results.
Mixed or inconsistent tone
If your TM includes very different types of content (for example, legal and marketing texts) in the same storage, the AI may mix their styles. You could end up with marketing copy that sounds like a contract, or legal text that reads too casually.
To avoid this, separate TMs for content with distinct tones or purposes.
Low-quality or corrupted data
Retrieval-augmented generation (RAG) reuses what it retrieves. If your TM contains typos, encoding issues, or outdated phrases, those problems may appear in AI output too.
Simply put: good data leads to good results; bad data repeats bad habits.
AI profiles with other sources of data
Your AI profiles can also use other data sources besides translation memory — namely, existing translations in your projects.
Before you begin: Prepare at least 500 keys as examples
To help the AI learn your style, make sure your project has at least 500 high-quality translations. These will be used as examples for context.
When creating a custom AI profile, you’ll be able to choose which examples to include: either keys with a specific tag or translations marked as reviewed.
Tagging keys
You can tag keys by opening your project, finding the needed key, and clicking the small label icon next to its name. Pick an existing tag or type a new one:
If you need to tag several keys at once, select them using the checkboxes and choose Tags: add/remove from the bulk actions menu:
Enter one or more tags and press Add.
Marking translations as reviewed
To mark translations as reviewed, click the glasses icon next to the translation.
You can also do this in bulk the same way, using Review: set/unset.
If the review feature isn’t available, go to More > Settings and make sure Reviewing is enabled. You’ll need the Manage project settings permission for that.
How to create a new AI profile
To create a custom AI profile that uses existing translations as context, go to Team settings > Custom AI profiles and click Create AI profile.
In the dialog, click Next to begin setup.
Providing basic information
You’ll need to provide some basic information:
Name your AI profile — give it a clear, descriptive name, for example Profile for marketing content.
Choose the source language — select the base language you want this profile to use as context.
Once you're ready, click Next.
Setting up data sources
Here you’ll define which translation keys the AI should use as examples. Make sure to select language pairs with clear, human-reviewed translations — at least 500 examples per language. The more you include, the better the results. Avoid using translations of low or inconsistent quality, as they can reduce accuracy.
Choose target languages — pick one or more target languages from the dropdown. These are the languages where the AI will apply your existing translation context.
Choose data source — where the AI will pull the examples from:
Tags — translations under specific tags. (Recommended for more flexibility.)
Reviewed status in project — translations marked as reviewed.
Specify data source
If you selected Tags, choose one or more tags to use when searching for translation keys. Only tags from projects in your current team will be available.
If you selected Reviewed status in project, choose one or more projects.
All reviewed translations from those projects (for the chosen target language) will be used as context.
When everything looks good, click Next, then choose one or more projects to activate this AI profile for. Remember that each project can only have one AI profile active at a time.
Click Activate to finish setup. Your new profile will appear in the list.
You can click the More icon next to its name at any time to edit, deactivate, or delete the profile.
Special notes and known limitations
One profile per project. A project can only use one AI profile at a time.
Branching support. If your project uses branching, the same profile applies across all branches.
Fallback behavior. If no profile is assigned to a project or language pair, Lokalise falls back to the default AI profile (no TM or existing translations passed as examples).
Style guide vs. TM and existing translations. When both a style guide and TM entries (or existing translations) are available for the same segment, the AI will always prioritize the TM entries or existing translations.
The style guide only comes into play when there’s no matching TM entry or existing translation, acting as a fallback to guide tone and phrasing.
Examples for AI. Currently, custom AI profiles can only use TM entries or existing translations as examples.
No entry-level selection for TM. You can't pick individual TM entries: all entries for the chosen language pairs are considered. Use a date filter if you want to exclude older ones.
Multiple TMs. If several translation memories are attached to a project, all of them will be used for context. Their priority follows the order in the project settings, where you can reorder them.