Token Economics

Token Economics

Token cost is a first-class metric in AgentCTX. Every operation is measured in tokens, and the platform is designed to minimize waste at every layer.

The Token Problem

Without context management, agents waste massive amounts of tokens:

Activity	Without AgentCTX	With AgentCTX	Savings
Tool descriptions (70 tools)	9,950 tokens	850 tokens	91%
Tool invocations (150 calls)	7,350 tokens	1,800 tokens	76%
File search + retrieval	16,925 tokens	1,290 tokens	92%
Documentation search	11,900 tokens	890 tokens	93%
Total session	46,125 tokens	4,830 tokens	90%

How AgentCTX Saves Tokens

1. Single Tool (~300 tokens)

Instead of exposing 70+ tool schemas (~140 tokens each), AgentCTX exposes one tool: the actx gateway. Agents query through CTX:

?t github          → discover tools (3 tokens)
>t github.issues.create title="Fix" → call tool (~25 tokens)

2. Content-Addressed Deduplication

Static content (tool schemas, docs, configs) is stored once and referenced by hash:

First session:  Store full schema → SHA-256 hash (CAS)
Next sessions:  Reference by hash → 3 tokens vs 140+ tokens

3. Context Compaction (CTX-ACCE)

The Adaptive Context Compaction Engine reduces context window usage by:

• Removing stale entries
• Merging redundant memories
• Archiving accessed-but-not-recently-used data

Enable with: actx start --compact

4. Memory Decay

The 5-layer memory system automatically expires low-value memories, keeping the active context focused and token-efficient.

Economy Tracking

The gateway’s economy middleware tracks costs in real-time:

interface CostReport {
  inputTokens: number;
  outputTokens: number;
  totalCost: number;       // USD
  provider: string;        // "openai", "anthropic", etc.
  model: string;           // "gpt-4", "claude-sonnet", etc.
}

Budget Enforcement

Set per-agent or per-session token budgets:

economy:
  budget:
    perSession: 100000      # Max tokens per session
    perDay: 500000           # Max tokens per day

The budget middleware (pipeline position 3) rejects requests that would exceed the budget.

ROI Model

AgentCTX’s 12-dimension inference pipeline ROI model projects:

Dimension	Value
Token cost savings	36% of ROI
Recovered productivity	28% of ROI
Autonomous uptime	20% of ROI
GPU infrastructure efficiency	16% of ROI
Per-developer annual value	~$43,831/year

Token cost is important, but it’s only 36% of the total value. The majority comes from productivity gains — agents that remember context, share knowledge, and don’t repeat work.

See Also

• Introduction — the token waste problem
• Memory & Persistence — how decay reduces bloat
• Gateway & Routing — economy middleware