What is AiGain?
AiGain is a behavioral control layer that sits between your organization and its AI usage. It controls how Claude thinks — not just what you ask it, but the cognitive mode it operates in when responding.
Every employee using AI today gets the same default behavior regardless of their role, their task, or what the company actually needs. AiGain fixes that.
The Problem It Solves
A support agent needs an AI that explores empathetically and asks clarifying questions. An engineer needs one that executes immediately and ships code. A researcher needs one that goes deep and surfaces trade-offs.
Without AiGain, all three get the same model behavior. Tokens burn. Quality suffers. The company has no visibility and no control.
How It Works
AiGain acts as a drop-in proxy for the Anthropic API. Your existing Claude integration points to AiGain instead of Anthropic directly. One line of code changes.
# Before
base_url = "https://api.anthropic.com"
# After
base_url = "https://aigain-production.up.railway.app/v1"
Every API call passes through AiGain. It looks up the team key, resolves the behavioral policy for that team, injects the right system prompt instructions, then forwards the request to Anthropic. The response comes back unchanged.
Behavioral Modes
BUILDExecute mode
Claude picks the best approach and implements it immediately. No alternatives presented. No thinking out loud. Output only. Ideal for engineering teams shipping code.
EXPLOREAnalysis mode
Claude covers multiple approaches, surfaces trade-offs, asks clarifying questions, and ends with decision points. Ideal for support, research, and strategy work.
The Control Hierarchy
Behavioral policy flows top-down, with each level able to override the one above:
Org Policy
Company-wide defaults. Every team starts here unless overridden. Set the floor and ceiling — min/max intensity, allowed modes, governance rules.
Team Policy
Each team gets its own behavioral preset. Engineering gets BUILD. Support gets EXPLORE. Research gets deep depth and high verbosity. One API key per team.
Individual Override
Senior engineers can unlock higher intensity. Specific roles can get tighter scope. Tracked and visible to admins.
What Gets Tracked
Token usage per team — see exactly where your AI budget is going.
Behavioral states — every policy change is logged with timestamp.
Active overrides — see which individuals are running outside team defaults.
Estimated savings — tokens saved by throttling teams that don't need max intensity.
The Three Controls
Every policy has three faders. These are not model parameters — they rewrite how the AI thinks at the system level.
Intensity — Drive · Effort
How hard Claude pushes. High intensity means direct, terse, decisive output. Low intensity means exploratory, tentative, shows reasoning steps. Most teams want 0.6–0.8 for production work.
Depth — Reasoning · Thinking
How deeply Claude analyzes before responding. High depth surfaces trade-offs, edge cases, and second-order effects. Low depth prioritizes speed and directness. Research teams: 0.8+. Support teams: 0.3–0.5.
Verbosity — Room · Voice
How much space Claude takes up. High verbosity means full explanations, context, and commentary. Low verbosity means output only — dead room. Engineering teams ship better with verbosity at 0.2–0.4.
Team Behavioral Profiles
Every team gets a different configuration. The differences are not subtle — they produce measurably different outputs on the same prompt.
| Team |
Mode |
Intensity |
Depth |
Verbosity |
Expected Output |
| Engineering |
BUILD |
0.95 |
0.55 |
0.08 |
Terse, ships code, ~200 tokens |
| Research |
EXPLORE |
0.45 |
0.95 |
0.85 |
Deep analysis, verbose, ~1500 tokens |
| Support |
EXPLORE |
0.30 |
0.35 |
0.88 |
Warm, thorough, lots of words |
| Sales |
BUILD |
0.82 |
0.22 |
0.42 |
Punchy, closes, no hedging |
| Legal |
BUILD |
0.88 |
0.90 |
0.10 |
Dense, precise, every word loads |
| Executive |
BUILD |
1.0 |
0.15 |
0.05 |
One answer. No reasoning. Max savings. |
Best demo: Compare Engineering vs Research on the same prompt. Expect 5–8× token difference — same task, radically different cost.
The Compare Engine
Run the same prompt through two different team policies simultaneously. AiGain scores both outputs across 6 metrics and tells you which behavioral configuration won — and why.
Metrics: Adherence, Depth, Clarity, Efficiency, Confidence, Token Efficiency. Each scored 0–100. A plain-English summary explains the behavioral difference.
Every run is stored. The Stats view builds a permanent record of which teams' behavioral states produce the best outputs for different task types.
Integration — 60 Seconds
Step 1: Create a team in AiGain. Get an API key.
Step 2: Point your SDK at AiGain instead of Anthropic.
client = anthropic.Anthropic(
api_key="your-aigain-team-key",
base_url="https://aigain.creativekonsoles.com/v1"
)
Step 3: Use Claude exactly as you do today. Every call is intercepted, the behavioral policy injected, then forwarded. Your code doesn't change — only how Claude thinks.
Security & Privacy
AiGain is a pass-through proxy. Your prompts and outputs are not stored. AiGain only logs token counts, behavioral state snapshots, and timestamps — never content.
API keys are hashed before storage. Team keys are scoped — a team key cannot read another team's policy or usage data. Your Anthropic API key lives only on the AiGain server, never exposed to clients.
Who Is This For
Engineering teams burning tokens on verbose AI output that slows down code review.
Support orgs that need AI to ask questions, not give answers.
Research teams that need AI to go deep, not fast.
CTOs and AI leads who need visibility and control over org-wide AI behavior — not just which model you're using.
If you're running more than 5 people on Claude, you need behavioral policy. AiGain is the layer that makes that possible.