False positiveflagged, but harmless
“I'll absolutely destroy you at chess tonight”
// caught on the word "destroy". Genuine banter, wrongly removed.
Coming soon • join our Discord
Moderate by meaning,
not by keywords.
Stridel enforces your community's policies across Discord, Reddit, Twitch and beyond. Tell it what a rule means to you, and it learns to recognize that, not just the words you used.
- Early access
- SOC2-ready
- <40ms decisions
Moderating on Discord · Reddit · Twitch · YouTube
The problem
Keyword triggers aren't enough.
Filters match strings. Communities communicate meaning. A word-list can't tell a joke from a threat, so it punishes the wrong people and misses the real harm.
False negativeallowed, but harmful
"The day you were born was a loss to humanity."
// no trigger words, so it sails through. Real harm, invisible.
You tune word-lists forever and still get it wrong. Stridel moderates the meaning instead ↓
How it works
Every message is a point in space.
Similar messages end up close together in a shared mathematical space. Your policy becomes a line through that space, and it shifts the moment you correct it.
Scroll to advance
policy boundary · "Keep a respectful tone"
friendly banterspoilersgraphic violenceharassment
Simplified 2D illustration. Real embeddings live in thousands of dimensions.
Capabilities
Control the boundary, not a word-list.
POLICY
Custom policies in plain language
Write a rule the way you'd explain it to a mod: "no graphic gore, but horror-game talk is fine." Stridel builds a boundary that means exactly that.REALTIME
Realtime decision-boundary updates
Report a message and the boundary moves instantly. No retraining, no waiting. The policy sharpens live as your moderators work.PLATFORMS
Multi-platform
Discord, Reddit, Twitch, and anything with a webhook. One policy, everywhere.FEEDBACK
Report-to-refine
Every moderator correction becomes a data point that tightens the line.LATENCY
Sub-second decisions
Fast enough to gate a message before it ever renders in chat.ISOLATION
Per-policy space
Each rule gets its own reshaped space, so policies never bleed into each other.Comparing to an LLM
Why not just an LLM?
A frontier LLM might edge us out by a few percentage points on the hardest edge cases. But it costs an order of magnitude more and burns far more power to get there. At scale, that trade rarely holds up.
A frontier LLMStridel
Accuracy on borderline casesA couple of percentage points on the hardest edge cases.
Slightly higher
Within a few points
Cost per 1k messagesRoughly 20× cheaper at volume, on the same call.
~$1–2
from $0.07
Latency per decisionFast enough to gate a message before most people read it.
> 5s
< 500ms
Energy per decisionA sliver of the compute results in meaningfully less electricity.
≈1000×
1×
Nearly the same accuracy at a fraction of the cost and carbon.The small gap isn't worth paying 20× for.
Pricing
Pricing at launch.
Free
$0/ forever
For a single community finding its feet.
- ◇3 policies
- ◇10k messages / mo
- ◇1 platform
- ◇Community support
Plus
$29/ mo
For active servers that moderate daily.
- ◇10 policies
- ◇250k messages / mo
- ◇1 platform
- ◇Report-to-refine
- ◇Email support
Most popular
Pro
$99/ mo
For networks running many communities.
- ◇Unlimited policies
- ◇2M messages / mo
- ◇All platforms
- ◇Realtime boundary updates
- ◇Analytics + priority support
Unlimited
$349/ mo
For platforms operating at scale.
- ◇Everything in Pro
- ◇10M messages / mo
- ◇Custom models
- ◇SSO + audit log
- ◇99.9% SLA
Pay-as-you-go API
Bring your own policies
The same latent moderation over a REST and streaming endpoint. Usage-based, volume-tiered.
0 – 1M msgs$0.20/ 1k
1M – 10M msgs$0.12/ 1k
10M+ msgs$0.07/ 1k
Coming soon
Draw your line.
We'll hold it.
Spin up a policy in minutes and watch Stridel enforce your community's meaning, not a word-list.