What Are Buyers?

Buyers are developers whose AI agents need specialized knowledge they don't currently have. The developer configures the system; the agent does the discovering and transacting.

How Buying Works

The buyer experience has two parts: the developer who configures and reviews, and the agent that discovers and transacts.

What the Developer Does

1. Instruments the agent - adds Kate tracing to record the agent's behavior
2. Configures discovery - sets budget limits, quality thresholds, and autonomy level
3. Reviews candidates - looks at discovered artifacts and approves subscriptions (or lets the agent act autonomously within configured limits)
4. Monitors improvement - tracks whether acquired knowledge is actually making the agent better

What the Agent Does (Autonomously)

1. Runs normally - handles user requests, generates outputs, uses tools
2. Gets evaluated - Kate scores runs and identifies where the agent is weak
3. Gaps are identified - compiled into a knowledge brief ("agent lacks SEO keyword research frameworks")
4. Discovery runs - finds artifacts that match the gaps
5. Subscribes - acquires access to relevant artifacts
6. Queries knowledge - when handling a request in a subscribed domain, queries the artifact for grounded answers

Discovery Configuration

Developers control how autonomous their agent's knowledge acquisition is:

config = await client.discovery.configure(
    agent_id=agent.id,
    mode="manual",                    # "manual" or "scheduled"
    max_tokens_per_action=500,        # Max tokens per discovery action
    daily_action_limit=3,             # Max actions per day
    min_compatibility_score=0.5,      # Minimum match quality (0-1)
    max_candidates_per_gap=3,         # Candidates to evaluate per gap
)

Knowledge Briefs

After enough traces are recorded, Kate compiles a knowledge brief for your agent - a summary of what the agent knows and what it's missing.

Briefs are versioned. You can compare versions to see how your agent's knowledge state has changed:

# Get the current brief
brief = await client.briefs.get(agent_id=agent.id)
print(brief.gap_summary)  # ["Lacks SEO keyword research frameworks", ...]

# Compare versions
diff = await client.briefs.diff(
    agent_id=agent.id,
    from_version="1",
    to_version="3",
)
print(diff.summary)

Improvement Tracking

After acquiring knowledge, Kate's evaluation system tracks whether the agent actually improves:

• Before acquisition: agent gives generic advice in the domain
• After acquisition: agent queries the artifact and gives specific, grounded answers
• Evaluation scores: Kate compares run scores before and after to quantify improvement

You can see this in the dashboard as score trends over time, or query it via the SDK:

summary = await client.evals.summary(agent_id=agent.id)
print(f"Overall score: {summary.overall_score}")
print(f"Regression detected: {summary.regression_detected}")
print(f"Recommendations: {summary.recommendations}")

Cost Controls

Buyers have several levers to control spending:

• Per-action budget - max_tokens_per_action caps individual transactions
• Daily limits - daily_action_limit prevents runaway spending
• Quality threshold - min_compatibility_score ensures you only pay for high-quality matches
• Manual review - set mode="manual" to approve every subscription personally
• Token wallet - your wallet balance is the hard cap; when it's empty, no transactions happen

Next Steps

• Buyer Quickstart - instrument and run your first discovery
• Configure Discovery - advanced settings
• Knowledge Briefs - understanding your agent's knowledge state
• Adoption Patterns - common buyer agent patterns