# Ternary Intelligence Stack

Ternary Intelligence Stack — an MCP server with 30 tools that hands your agent a third answer: hold, meaning the evidence is thin and more should be gathered before it acts. No key required.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · 3210 installs
- Author: rfi-irfos
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://smithery.ai/server/rfi-irfos/ternlang
- Updated: 2026-04-17
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/mcp/ternary-intelligence-stack

## What it does

Your agent normally chooses between two answers. This server adds a third. `hold` says the evidence is not enough yet and the agent should collect more before committing. Verdicts come out of numbers: `trit_decide` converts a float into a trit with a confidence score, `trit_vector` averages named dimensions by weight, and `trit_action_gate` vetoes an action as soon as a dimension flagged `hard_block:true` carries negative evidence.

The tool list goes past routing. `moe_orchestrate` sends a query to 13 specialist agents and keeps a safety veto at the end, `ternlang_run` compiles and runs `.tern` programs on the BET VM, `trit_audit` scans a decision log for forced yes/no calls and scores it against Article 13 and Article 14 of the EU AI Act. Three memory layers with TTL and a consolidation cycle sit alongside.

Connect over HTTP to `https://ternlang--rfi-irfos.run.tools`. Nothing to pay to start. The free `moe_orchestrate` preview returns the verdict plus the voices of three experts only. The whole 13-expert panel, SSE streaming and server-side memory need a paid tier: Tier 2 at €24.99/mo for 10k calls or Tier 3 at €49.99/mo for 20k.

## Requirements
- Remote server: `https://ternlang--rfi-irfos.run.tools`

## Install

### Claude Code (native platform)

**Platform CLI: remote server**
```bash
claude mcp add --transport http ternlang https://ternlang--rfi-irfos.run.tools
```

### Codex

**Platform CLI: remote server**
```bash
codex mcp add ternary-intelligence-stack --url https://ternlang--rfi-irfos.run.tools
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "ternary-intelligence-stack": {
            "url": "https://ternlang--rfi-irfos.run.tools"
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "ternary-intelligence-stack": {
            "type": "http",
            "url": "https://ternlang--rfi-irfos.run.tools"
        }
    }
}
```

### Gemini CLI

**Platform CLI: remote server**
```bash
gemini mcp add -t http ternary-intelligence-stack https://ternlang--rfi-irfos.run.tools
```

## When to use it
- turn several independent scores into one verdict with a confidence value
- veto an action when a dimension is marked `hard_block`
- run two conflicting claims past 13 experts and get a synthesis
- spot forced yes/no calls in an agent's decision log
- rewrite `if/elif/else` branches as `.tern` with explicit hold arms

## Example prompts
- Call `trit_decide` on 0.62 and 0.58 and tell me whether confidence is high enough to act
- Audit the agent's decision log with `trit_audit` and list the calls that should have been held
- Aggregate three weighted dimensions with `trit_vector` and check if the action is actionable

## Similar
- [constitutional-ai](https://affpapa.org/en/nevibecode/skills/constitutional-ai.md) — skill · 1100 installs
- [anthropic-technical-deep-dive](https://affpapa.org/en/nevibecode/skills/anthropic-technical-deep-dive.md) — skill · 141 installs
- [humane-proxy](https://affpapa.org/en/nevibecode/mcp/humane-proxy.md) — MCP server · ★ 27 repo stars
- [ai-prompt-engineering-safety-review](https://affpapa.org/en/nevibecode/skills/ai-prompt-engineering-safety-review.md) — skill · 10327 installs
- [hallucination-investigator](https://affpapa.org/en/nevibecode/agents/hallucination-investigator.md) — subagent · ★ 6505 repo stars
- [model-risk-manager](https://affpapa.org/en/nevibecode/agents/model-risk-manager.md) — subagent · ★ 6505 repo stars

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
