Skip to main content

Using Toby

Toby is ViralToby's autonomous AI agent. Unlike a simple scheduler or template system, Toby is a learning system that creates content, analyzes performance, discovers trends, and continuously improves its strategy based on what works for your audience.

What Toby Is

Toby is an autonomous content creation agent that:

  • Generates original content (Reels, Carousels, Threads) based on your Content DNA
  • Learns which strategies perform best using Thompson Sampling
  • Maintains your content buffer so you always have posts ready
  • Scores published content and feeds results back into its strategy
  • Discovers trending topics and adapts your content plan
  • Operates continuously in the background without manual intervention

Toby is not a set-it-and-forget-it tool. It actively evolves its approach over time, getting better at creating content that resonates with your specific audience.


Enabling and Disabling Toby

Enabling Toby

Toby can be enabled or disabled per brand from your brand settings. When you enable Toby:

  1. Toby starts in Bootstrap phase (aggressive content generation to fill your buffer).
  2. It immediately begins checking your content buffer and generating new content.
  3. Generated content appears in your Pipeline for review.

Disabling Toby

When you disable Toby:

  • Content generation stops.
  • Already-scheduled content remains scheduled and will still publish.
  • Your strategy scores and learning data are preserved -- if you re-enable Toby, it picks up where it left off.

The Tick Loop

Toby runs on a 5-minute tick loop. Every 5 minutes, the orchestrator wakes up and processes all users with Toby enabled.

What Happens Each Tick

On each tick, Toby evaluates the following actions in priority order and executes the most needed one:

PriorityActionWhat It Does
0Quality GuardSelf-monitoring sweep: detects and cancels duplicate titles, caption duplicates, slot collisions, and fallback content in your scheduled queue.
1Buffer CheckChecks if all content slots for the upcoming days are filled. If not, generates new content to fill the gaps.
2Metrics CheckLooks for published posts older than 48 hours that have not been scored yet. Fetches engagement metrics and scores them. (Runs every 6 hours.)
3Analysis CheckUpdates strategy scores based on new metrics. Detects performance drift and adjusts explore ratio if needed. (Runs every 6 hours.)
4Discovery CheckTriggers a trend scout scan to identify emerging topics and opportunities in your niche.
5Phase CheckEvaluates whether Toby should transition to the next learning phase.
info

Each step runs independently. If a buffer check fails, it does not prevent the metrics check from running on the next tick. This prevents cascade failures where one error blocks all Toby activity.


Three Learning Phases

Toby progresses through three phases as it accumulates performance data:

Phase 1: Bootstrap

Goal: Fill your content buffer as quickly as possible.

  • Generation rate: Up to 6 pieces per brand per hour (aggressive).
  • Cooldown between generations: 2 minutes per brand.
  • Strategy selection: Explores all options broadly with minimal preference.
  • When it transitions: After the buffer is full and initial metrics start arriving.

Bootstrap mode is designed for new accounts. Toby generates content rapidly to ensure you have a full pipeline of content ready for review.

Phase 2: Learning

Goal: Gather enough data to identify winning strategies.

  • Generation rate: Up to 2 pieces per brand per hour (steady).
  • Cooldown between generations: 15 minutes per brand.
  • Strategy selection: Starts favoring strategies with positive early signals while still exploring broadly.
  • When it transitions: After sufficient performance data has accumulated across multiple strategy dimensions.

Phase 3: Optimizing

Goal: Maximize engagement by exploiting proven strategies while continuing to explore.

  • Generation rate: Same as Learning (2 per brand per hour).
  • Strategy selection: Full Thompson Sampling -- strategies are selected probabilistically based on their performance distributions. Proven winners get higher probability, but exploration never stops.
  • Ongoing: This is the steady-state phase. Toby continues optimizing indefinitely.

Content Buffer

Toby maintains a 7-day content buffer for each brand. This means:

  • You always have 7 days of content ready to go.
  • If you approve content slower than Toby generates it, the buffer grows.
  • If the buffer starts running low (content is being published faster than generated), Toby increases generation priority.
  • You receive email reminders if your buffer drops below safe levels.

The buffer ensures you never run out of content, even if you do not log in for a week.


Thompson Sampling Explained

Thompson Sampling is the algorithm at the heart of Toby's learning. Here is how it works in simple terms:

The Problem

Toby has many strategy options for each content dimension (personality, topic, hook style, title format). It needs to figure out which combinations work best for your audience -- but it also cannot stop trying new things, because what works today might not work tomorrow.

How It Works

  1. Each strategy option has a performance distribution. Think of it as a range from "probably bad" to "probably great," with varying confidence.

  2. To choose a strategy, Toby samples from each option's distribution. It randomly picks a value from each distribution, then uses whichever option produced the highest sample.

  3. After seeing results, Toby updates the distributions. Good performance narrows the distribution toward higher values. Poor performance narrows it toward lower values.

  4. The randomness is the key. A strategy that Toby thinks is great (narrow distribution around a high value) will be selected most of the time. But a strategy with less data (wide distribution) has a chance of producing a high sample -- that is the exploration.

Why It Works

  • Exploitation: Proven strategies are selected more often because their distributions are centered on higher values.
  • Exploration: Unproven or new strategies can still win the sample, ensuring Toby tries them periodically.
  • Automatic balance: The algorithm naturally shifts from exploration to exploitation as more data arrives. No manual tuning needed.
tip

You do not need to understand the math behind Thompson Sampling. The takeaway is: Toby tries different approaches, learns which ones get better engagement, and gradually shifts toward winners -- while keeping a healthy exploration rate to discover new opportunities.


Cognitive Agents

Toby is not a single monolithic AI. It is composed of specialized cognitive agents, each responsible for a different aspect of the content creation and learning process:

AgentRole
CreatorGenerates content using memory-augmented prompts. Retrieves relevant memories and strategy context to produce content aligned with what works.
CriticMulti-critic ensemble that evaluates content quality. Combines rule-based checks, semantic analysis, and audience simulation to score content before it enters the pipeline.
AnalystEnhanced scoring engine with anomaly detection. Identifies breakout content, detects unusual performance patterns, and performs causal attribution.
StrategistChain-of-thought strategy reasoning. Uses deep reasoning to select the best strategy combinations, taking into account performance data, trends, and context.
ReflectorTriple memory writer. After content is scored, the Reflector writes observations to Toby's episodic, semantic, and procedural memory systems.
ScoutEnvironmental context gatherer. Collects signals from memory, world model, and trend data to provide context for the Strategist and Creator.
IntelligenceContinuous signal processing. Gathers and processes intelligence from various sources to keep Toby aware of the broader landscape.

Additional supporting agents:

AgentRole
Quality GuardSelf-monitoring agent that sweeps the scheduled queue for duplicates, collisions, and low-quality content.
Meta LearnerHigher-order learning agent that adjusts Toby's learning parameters (like explore ratio) based on meta-level observations.
Pattern AnalyzerIdentifies recurring patterns in content performance across multiple dimensions.
Experiment DesignerDesigns A/B experiments to systematically test strategy hypotheses.

Toby's Memory System

Toby has a structured memory system inspired by how human memory works:

Episodic Memory

Records of specific events and outcomes. "Post X was published on Monday and got 2x the average saves." These are concrete, timestamped memories that the Creator and Analyst reference.

Semantic Memory

General knowledge extracted from patterns. "Myth-busting hooks perform 40% better than listicle hooks for this DNA." These are abstracted insights that inform strategy selection.

Procedural Memory

Learned rules and heuristics. "When creating content about nutrition, always include a specific study citation." These are operational rules that guide content creation.

Memory Gardening

Toby's memory is not unlimited. A background process called the Memory Gardener periodically prunes outdated or low-value memories, consolidates redundant entries, and maintains the health of the memory system.


Explore Ratio

The explore ratio controls how much Toby experiments versus sticking with proven strategies:

  • Default: 35% of content explores new strategies
  • 65% exploits proven strategies

This ratio is not static. Toby's Meta Learner can adjust it based on conditions:

  • If performance is declining, the explore ratio may increase to discover new winning strategies.
  • If performance is stable and strong, the ratio may decrease to double down on what works.
  • Drift detection runs periodically and adjusts the ratio if the audience's preferences seem to be changing.

Activity Log

You can view Toby's activity log from your dashboard. The log shows:

  • When Toby generated content and for which brand
  • Buffer check results (how many slots were filled)
  • Metrics scoring events (which posts were scored, how they performed)
  • Strategy analysis updates
  • Quality Guard actions (duplicates cancelled, slots repositioned)
  • Phase transitions
  • Any errors Toby encountered

The activity log is your window into what Toby is doing and why.


Budget Management

Toby tracks AI generation costs per user:

  • DeepSeek API costs -- Text generation tokens (input and output)
  • Image generation costs -- Per-image pricing for AI backgrounds
  • Daily cost records are maintained and visible in your settings

You can set a daily budget (in cents) to cap Toby's spending. When the budget is reached, Toby pauses generation until the next day.


Rate Limits

To prevent over-generation and ensure stable operation, Toby enforces rate limits:

LimitNormal ModeBootstrap Mode
Max per brand per hour26
Max per user per hour620
Cooldown between same-brand generations15 minutes2 minutes

These limits prevent Toby from generating excessive content during a single tick, keeping costs predictable and quality high.


Tips for Getting the Most Out of Toby

1. Write a detailed Content DNA

The more specific and thorough your Content DNA, the better Toby's output from day one. Include examples, specific tone words, and clear topic boundaries.

2. Review content in the Pipeline

Do not let content pile up in "Pending Review." Timely approvals and rejections give Toby faster feedback, which accelerates learning.

3. Be patient with the learning curve

Toby needs real performance data to optimize. The first week or two will be exploration-heavy. By week three or four, you should see noticeably improved content selection.

4. Check the activity log

If Toby seems to be generating content you do not like, check the activity log to understand its reasoning. Then adjust your Content DNA accordingly.

5. Do not change your DNA too frequently

Give Toby time to learn within a stable DNA. Constant DNA changes reset the learning process. Make incremental refinements rather than wholesale rewrites.

6. Keep your platforms connected

If a platform token expires or a connection drops, Toby cannot publish to that platform. Check your connection status periodically in brand settings.