Changelog

Version 1.7.1

Released: 2025-12-10

Fixed

Fixed AI Policies not working after a recent update.

Version 1.7.0

Released: 2025-12-05

Added

We now have a ‘Feedback Form’ that shows up when you click the ⚠️ button next to the version on the bottom-right corner of the webapp.
Moderation page: Semantic Search for AI Policies (to test if they’d match one of the examples);
Moderation page: Text Search has been improved to also test the RegEx’s if any exist.

Fixed

‘Test AI Policies’ has been fixed.
KB Entry questions testing has been fixed as well.

Version 1.6.4

Released: 2025-11-28

Added

You can now test AI Policies and Pattern Rules, similarly to how you can execute a ‘Semantic Search’ in the Knowledge Base page;
Created a new ‘Calibration Mode’ that allows you to turn on AI Policies and Pattern Rules (including their punishment actions), but without actually executing the punishments (only logging that it would’ve done so if the calibration mode was off); this is often called doing a “dry run” in IT terms/lingo.

Fixed

After v1.6.1, the AI would often find much more Knowledge Entries (which is good in most cases), but sometimes even irrelevant ones. This has been fixed by adding an evaluation step that filters out irrelevant results; reducing significantly the number of “false positives” (AI responding when it shouldn’t have done so).

Version 1.6.3

Released: 2025-11-24

Added

‘Draft Management’ in the webapp, local unsubmitted changes are now locally saved automatically, prompting the user to ‘continue editing’ their work or ‘discard unsaved changes’ when navigating away from the page. This ensures that users don’t lose their progress and can easily resume where they left off.

Version 1.6.2

Released: 2025-11-21

Fixed

In some cases inside our webapp, when you click an opened/focused text field, it would undo text editing, causing frustration. This has now been resolved.

Version 1.6.1

Released: 2025-11-21

Added

The Community Self-Reporting feature now also includes a response to the offender (if the offender hasn’t been banned, in which case that wouldn’t matter); this instructs the user why an action (or lack thereof) was taken.
New internal feature for ‘sanitizing user questions’ — this increases the level of matches from the knowledge base, specially if the user asks something super long (or with multiple questions).
- Unfortunately, however, this has made the AI too ‘trigger happy’, so we’re adding yet another step (for paid plans) to evaluate the KB outputs and filter out irrelevant results to the user’s original question (if all results are irrelevant, the AI will not answer, to avoid spamming the chat when unprompted).
- Edit: Fixed in 1.6.4.

Version 1.6.0

Released: 2025-11-19

Added

You can now add more than one social account to the same login, so you can sign-in with either accounts which are most easily accessible to you. To do this, head on over to your personal account page (click your profile picture on the top-right corner then select ‘Profile’ from the menu). You can also unlink accounts from there.
Note: For now we can’t add a Telegram login to an existing account, but if you create an account with Telegram, you can then add other social accounts to it.

Version 1.5.0

Released: 2025-11-18

Added

Logs for everyone 🪵! Under the Organization settings, you can now add a Logs group on Telegram or select an existing Discord channel. These logs inform you in real-time of detected unanswered questions (which you can also see under your organization’s dashboard), usage alerts, billing issues, and more. We are planning to split these logs into separate categories so you can filter by them or enable/disable notifications based on each category,etc.;
We also added some internal logs for our own awareness of important events happening on our platform, so we can best serve our customers! This is also in preparation for a new ‘Submit Feedback’ form we’ll introduce at a later version.

Fixed

Some billing querks around paid-plan-only features (AI Policies, Custom Personalities, Demo community & Connection counts)

Version 1.4.0

Released: 2025-11-07

This release doesn’t include any user-facing changes; it’s just in preparation for the next larger release (1.5.0), and we’ll include some minor fixes in this version soon too.

Fixed

We rebuilt a large portion of our backend API to prepare it for supporting streaming endpoints (better UX from slower inference providers, Telegram even supports streaming natively! so expect that to come to our TG bot too soon);
We will be adding better ‘agentic capabilities’ to C12s in the next upcoming releases (thanks to this refactor), allowing it to interact with other systems and services more intelligently with higher contextual awareness;
We now trace embeddings as well (this gives us better visibility into the embedding model/provider time taken; we can benchmark different providers and models better, etc.);

Version 1.3.3

Released: 2025-11-06

Added

Unanswered Questions on the dashboard now have a column that indicates if they were sent on Discord or Telegram; on hover, they will display the name of the group/server and channel;

Fixed

The ‘go to’ button [↗️] on the dashboard now correctly navigates to the corresponding message (Discord messages weren’t being linked properly at all, and Telegram ones weren’t being formatted correctly for public groups).

Version 1.3.2

Released: 2025-11-03

Added

Discord bot → Added a new ‘Try again without sources.’ button when the AI can’t find relevant Knowledge Base entries. This button will try asking the model directly (prone to hallucinations unless the question is something the LLM knows intrinsically);
Discord bot → (feature-flagged for now, let us know if you’d like to try it): New ‘raw_llm’ command to execute “raw prompts” to the LLM; useful when testing custom endpoints/models.
BYOK support → We added backend support for ‘Bring Your Own Keys’ (BYOK), so you can use your own self-hosted/finetuned models with our platform. BYOK is under closed Beta, please contact us if this is important for you.

Fixed

Fixed issues found with our Discord bot’s writing “thread mode”;
AI responses larger than 2000 characters are now split at the nearest paragraph boundary, so large messages are sent in multiple parts.
Billing page → Fixed an issue where the billing stats for ‘Analysed Messages’ was not displaying correctly.

Version 1.3.1

Released: 2025-10-25

Added

Stats → Community Growth: Added a view toggle to switch between Last 24 hours and Last 7 days graphs.
Stats → Community Growth: Simplified the tooltip—platforms with 0 values are hidden.

Fixed

Resolved an issue where Discord channel permissions overlapped with their parent category, causing unexpected access errors.
Fixed a bug in Stats where admin messages were being counted twice.

Version 1.3.0

Released: 2025-10-23

Added

New Statistics graph: Community Growth (Joins/Leaves)
Community Engagement graph now has a toggle for WAU’s (weekly active users);

Fixed

Webapp colors now match the logo;
“Report feature” renamed to “Community Self-Report” and now links to the relevant documentation page;

Version 1.2.0

Released: 2025-10-21

Added

Added new content moderation pattern rules (new content-types for user-bot messages and system messages).
PostHog integration to properly attribute referral websites to signed-up customers (laying the ground work for more advanced internal analytics and eventually creating a referral program).

Fixed

The ‘Community Engagement’ graph now works properly (tracking DAUs on Discord and Telegram). More updates are coming to this section.
The ‘Community Activity’ graph now displays gray squares for hours in the future (when no data is available), with a nice tooltip.
Removed the trend lines on the graphs until they are functional.
Improved login/signup error handling.
Regular Expressions now automatically include the ‘m’ flag (multiline).
The bot now executes the chosen actions when a user /reports a message (previously it was only logging the report).
The ‘demo mode’ integration (+info here) now properly does not count towards plans connection limits (use it for testing!).

Version 1.1.1

Released: 2025-10-17

Fixed

We introduced a regression with the last ‘mobile refresh’ update, which caused issues in the ‘Unanswered Questions’ dashboard when a question was too large (without any spaces, such as a long URL). This has now been fixed.

Version 1.1.0

Released: 2025-10-12

Added

New badge on the bottom right corner with the current version number.
New feature: Community self-moderation via the /report command (disabled by default).

Fixed

We made our webapp generally more mobile-friendly. While a few minor features remain ‘desktop-only’, most of the app should now work seamlessly on mobile devices.
Added the same SEO as our homepage to the webapp, so our webapp links (specially the invites) are prettier when shared.

Version 1.0.0

Released: 2025-10-07

This release marks the initial release of C12s, finally opening the door to the public, after over a year of development behind closed doors with several large teams.

Added

New section “Moderation Policies & Rules”; tip: check the spam protection article.

Fixed

In the statistics section, most graphs stopped working a few weeks ago due to a backend migration. This has now been fixed and the first and last graph are now functioning properly.
Because registrations are now open, organization invites now work properly without requiring any intervention from our side.
We switched authentication systems (from a 3rd-party provider to our own using BetterAuth’s library).