AMIS analyses your mod queue in real time, classifies posts against your subreddit's own rules using a two-tier ML pipeline, and surfaces prioritised recommendations — all without leaving Reddit.
Built natively on Devvit — no external dashboards, no browser extensions, no context switching.
Every new post is classified within seconds via a PostSubmit trigger. Configurable auto-remove/approve thresholds let AMIS act without moderator intervention.
Cheap vector embedding search pre-screens posts. LLM inference runs only for ambiguous cases (0.45–0.85 similarity range), reducing AI costs by 60–70%.
Ingests your subreddit's rules, automod YAML, wiki pages, and removal reasons into a semantic embedding store. Every recommendation is grounded in your rules.
Cosine similarity match produces a suggested action (remove / approve / escalate / monitor), a confidence percentage, and a plain-English AI rationale.
Approve verdicts are cached for 24 hours; non-approve verdicts cached until the post changes. 7-day staleness filter prevents re-analysis of old content.
Paginated, filterable log of all moderation actions with AND-logic filters: By AI, Approved, Removed, and Escalated.
Analyses 50+ mod decisions for enforcement drift, divergence, and rule-category variance. Surfaces insights directly in the dashboard.
The custom post auto-refreshes every 15 seconds. No manual page reload needed. Live countdown banners show background analysis progress.
Near-duplicate posts (>85% embedding similarity) are flagged automatically, helping you catch spam and repost waves early.
AMIS installs directly into Reddit as a Devvit app — no servers to manage.
Go to your subreddit → Mod Tools → App Directory → find AMIS → Install. Or use the Reddit app link below.
In the Reddit Developer Portal, open App Settings and enter your AI provider API key (OpenAI, Gemini, Claude, or custom).
Run AMIS: Refresh Policy from Mod Tools. This reads your subreddit rules, automod YAML, and wiki pages into the embedding store.
Run AMIS: Open Dashboard. The dashboard auto-polls every 15 seconds and surfaces the most urgent posts for review.
Use Remove, Approve, Skip, or Escalate buttons. Every action is logged permanently in the History view.
Weekly, run AMIS: Consistency Check to detect enforcement drift and review the 💡 Insights panel for analytics.
| Menu Label | What It Does | When to Run |
|---|---|---|
| 🏥 AMIS: Health Check | Shows item count, embedding count, and pending insight count | After install; verify setup |
| 📜 AMIS: Refresh Policy | Re-ingests subreddit rules, automod YAML, wiki pages, removal reasons | After changing any rules |
| 🔍 AMIS: Analyze Queue | Schedules a background batch analysis of the mod queue | On demand (also runs every 2 min) |
| 📊 AMIS: Consistency Check | Runs consistency analytics (requires ≥50 mod decisions) | Weekly or after high-volume activity |
| 🖥️ AMIS: Open Dashboard | Creates or navigates to the AMIS dashboard post | Daily triage |
AMIS uses consistent iconography across the dashboard. Here's what everything means.
Opens the paginated log of all past moderation actions.
Consistency analytics, enforcement drift alerts, and rule-category variance reports.
AI provider info, full-screen layout toggle, and manual queue refresh.
Shows a live countdown (e.g. "⏳ AI analysing — 38s") while background analysis runs.
Immediately calls reddit.approve() on the current post. Action is permanent and logged.
Immediately calls reddit.remove() on the current post. Action is permanent and logged.
Moves to the next post without taking action. Post remains in queue.
Marks the post for senior moderator review. Logged in History as Escalated.
Displays embedding count, item count, and pending insight count for diagnostics.
Manually triggers a fresh queue analysis. Also runs automatically every 2 minutes.
AI confidence score (0–100%) shown on every recommendation. Higher = more certain.
Indicates the action was taken by AMIS automatically (auto-remove/approve), not a human.
One-at-a-time review of flagged posts with Remove / Approve / Skip / Escalate actions.
Paginated log of all actions. Filter by: By AI, Approved, Removed, Escalated (AND logic).
Consistency analytics, enforcement drift, and rule-variance analysis from 50+ decisions.
AI provider details, full-screen mode toggle, and manual refresh controls.
AMIS supports multiple AI providers for text generation and embeddings.
| Setting | Required | Description |
|---|---|---|
| AI_PROVIDER | Optional | openai · gemini · claude · custom (default: openai) |
| AI_API_KEY | Required | API key for your chosen text-generation provider |
| VOYAGE_API_KEY | Claude/custom only | Voyage AI key for embeddings when using Claude or custom provider |
| CUSTOM_API_BASE_URL | custom only | OpenAI-compatible base URL (e.g. http://localhost:11434/v1) |
| AUTO_ACT_ENABLED | Optional | Enable automatic remove/approve on new posts (default: false) |
| AUTO_REMOVE_THRESHOLD | Optional | Confidence threshold for auto-remove 0–100 (default: 90) |
Install directly from the Reddit App Directory — free to use with your own AI API key.