AI Moderation You Can Trust
Spin up agents that read every comment, surface the hard calls, and reply in your voice. Every action ships with a confidence score and the reasoning behind it. Launch new agents in dry-run when you want to watch first.
Six things every agent ships with
Ship with the brakes fully on.
Every agent can optionally launch in dry-run. Every destructive action - bans, mass deletes, IP blocks - can be gated behind a human approval queue. The agent only sees the tools you opt into. The rest don't exist as far as the model is concerned.
- Disabled, dry-run, or enabled. Flip status without redeploying a prompt.
- Per-agent tool allowlist. Disallowed tools are hidden from the model entirely, not just refused at runtime.
- Granular ban powers. Opt in separately to deleting a banned user's history or banning by IP.
- Approvals queue with optional per-reviewer routing. Email the right person, not the whole team.
- Models from providers that don't train on your data.
| Agent | Proposed action | Target | Conf. | Review |
|---|---|---|---|---|
| ToS Enforcer | Ban user (IP) | @throwaway-9281, 4 prior flags | 0.97 | |
| Greeter | Warn via DM | First comment, off-topic on /forums/intro | 0.81 | |
| Spam Watcher | Mark spam, delete history | 47 outbound links in 9 minutes | 0.99 |
Decisions made with the whole picture.
An agent that only sees one comment will misjudge it. FastComments agents pull in the full thread, the commenter's history, the page they're commenting on, and any community guidelines you've written - then act on what they actually understand.
- Parent comment and prior replies in the same thread.
- Trust factor, account age, ban history, and recent comments for the commenter.
- Page title, subtitle, description, and meta tags.
- Tenant community guidelines prepended to every prompt for consistent rule application.
Wake on the right events. Take only the right actions.
Pick the events that should wake an agent up. Pick the tools it's allowed to use when it does. Both lists are explicit, both are enforced. Optional delays let votes and reports settle before the agent looks.
Triggers
- New comment posted
- User's first comment
- Flag count crosses threshold
- Vote threshold crossed
- Auto-captured by spam engine
- Moderator reviews a comment
- Comment edited or deleted
- Comment pinned, unpinned, locked, unlocked
- Optional delay up to 30 days before the agent runs
Actions
- Mark reviewed
- Approve / unapprove
- Mark spam
- Pin / unpin / lock / unlock
- Vote on a comment
- Write a comment
- Award a badge
- Warn a user via DM
- Ban a user
- Send an email
- Save a memory note
- 1Trigger
- 2Context
- 3Agent
- 4Action
- 5Audit
No invoice surprises.
Hard daily and monthly caps per agent and per account. Email alerts at every threshold you choose. When an agent hits its cap it pauses. No overage, no manual intervention.
- Daily and monthly caps on each individual agent.
- Account-wide rollups so one agent can't starve another.
- Custom alert thresholds. Default is 80% and 100%; we recommend adding 50% for noisy agents.
- Built-in rate limiting and concurrency caps with backpressure when the queue saturates.
- Skipped triggers are logged with the reason, so you know what you missed.
- Over agent budget
- Over account budget
- Rate-limited
- Concurrency saturated
- 50% alert sent
- 80% alert sent
- 100% - paused until next month
- Agent ToS Enforcer proposed Mark spam on c#9b1f8 - confidence 0.97
- Reasoning: "Comment is verbatim copy of a known affiliate spam template; user account 8 minutes old."
- Auto-executed (under approval threshold). Cost $0.0021.
- Agent ToS Enforcer proposed Ban user on @throwaway-9281
- Approved by devon@fastcomments.com. Action executed.
Built for publisher-grade compliance.
Every agent action is recorded with the reasoning, confidence, transcript, cost, and reviewer. Every full ban in the EU region requires a human in the loop, locked on by us, not by you.
- Per-run records: prompt, transcript, tools used, action taken, justification, reviewer.
- EU DSA Article 17: full bans require human approval in the EU region. Locked on; cannot be turned off.
- Tenant community guidelines drive moderation decisions. Your rules, your wording.
- Display names you control, distinguishing agent-authored comments from human-authored ones.
- Dry-run vs live counts in analytics so you can prove what's been automated and what hasn't.
Start from a template. Validate against your own data.
Spin a new agent up from a prebuilt template, then replay it against weeks of your historical comments to see what it would have done before turning it on.
- Prebuilt templates for common moderation use cases. Created agents start in dry-run.
- Replay against the last 24 hours, 7 days, 30 days, or any custom window up to 90 days.
- Cost estimate before you start; hard cap aborts the run if exceeded.
- Replay results split into matches, deltas, no-action, and errors so you can focus on the disagreements.
- All1,284
- Matches1,031
- Deltas148
- No action97
- Errors8
The boring parts, handled.
- 01
Per-comment execution locking. Two agents never trip over each other on the same comment.
- 02
Loop prevention so one agent's reply doesn't endlessly trigger another.
- 03
Failed runs retry with bounded attempts before being marked errored. Never lost.
- 04
Model choice per agent. Swap between cost-optimized and quality-optimized without rewriting prompts.
Join Our Happy Customers
AI agents are part of the FastComments platform. Sign up free, switch any agent to dry-run, watch what it would do, then go live.
Sign up free
.png "www.oliveoiltimes.com")
.png "learn.oliveoilschool.org")
.jpeg "jagunbae.com")
.png "aquaverse.fr"){kind=logo}
Frequently Asked Questions
We've compiled answers to some of the most common questions to help you get started.!
How does dry-run mode work?
An agent in dry-run runs end-to-end against every matching trigger. It picks tools, generates reasoning, decides on an action, but the action is never executed. Every run is logged so you can review what it would have done before flipping it to enabled.
Which models can I choose?
Today, GLM 5.1 (smarter, slightly slower) and GPT-OSS 120B Turbo (faster). Both are hosted by providers that don't train on your data. We add new models as the price-performance frontier moves.
How do budget caps work?
Each agent has its own daily and monthly cap, and your account has a rollup. When an agent crosses its cap, it pauses; alerts go out to billing admins at every percentage you've configured. Skipped triggers stay logged so you know exactly what was missed.
How are you DSA-ready?
EU DSA Article 17 requires human review of user suspensions. In the EU region, the 'ban' tool is locked behind the approval queue and cannot be fully automated. It's a platform-level rule we enforce, not a setting you can disable. Every action is recorded with reasoning and reviewer for the audit trail.
Can I test an agent before turning it on?
Yes. The replay feature runs an agent against your historical comments, up to 90 days back, and shows you matches, deltas, and disagreements vs the prior outcome. You see the cost estimate before starting and can set a hard cap that aborts the run if it overshoots.
Can I bring my own prompt and rules?
Each agent has an initial prompt that defines its personality and goals. You can also paste your community guidelines into the agent's context, and FastComments prepends them to every run so the agent applies your rules consistently.
Stand up your first agent today.
Pick a template, point it at the triggers you care about, and turn it on when you're ready. No credit card required to sign up.