Skip to main content
Glama

Project Christina: Autonomous Cognitive Amplifier

Python 3.11+ License: MIT Dependencies Tests FastMCP Compliant Code Style: Ponytail Ultra

Host: Google Antigravity Host: OpenCode Host: Claude Desktop Host: OpenAI Codex CLI

Quick Start • Architecture • MCP Tools • CLI Reference • Documentation • Contributing


🌟 Executive Overview: Eliminating Context Rot

Modern AI coding agents fail on large codebases ($10\text{M}+$ tokens) not due to reasoning limitations, but due to Context Rot: attention distraction, unindexed prompt pollution, lost architectural constraints across context wipes, and uncoordinated swarm collisions.

Project Christina solves this through an integrated, high-performance cognitive trinity:

  1. RLM (Recursive Language Models — Context as a Variable): Moves massive codebases out of conversational prompt transcripts into sandboxed heap variables (context). Agents execute bounded sub-queries ($D \le 2, N \le 20$) and transactional copy-on-write (CoW) Python control loops with automated rollback.

  2. Graphify (Deterministic CPG & AST Radar): Builds an in-memory SQLite3 WAL call graph delivering sub-0.3ms recursive CTE queries, transitive blast radius calculation, and God node centrality ranking with a 71.5x token discovery reduction.

  3. OKF (Durable Semantic Memory & Epistemic Evolution): Sub-microsecond regex frontmatter parsing ($6.72\ \mu\text{s}$/doc), formal AGM belief revision, Ebbinghaus decay, and Mark-and-Sweep epistemic garbage collection with an automated $\le 400$-token post-compaction memory recovery anchor.

  4. Cross-Project Epistemic Distillation: Anonymizes proprietary code into mathematical archetypes with a strict Zero-Information Leakage invariant ($\mathcal{I} = 0$), multi-base Shannon entropy secret scrubbing, and Bayesian invariant aggregation across code repositories.

  5. Swarm Concurrency Control: Decentralized multi-agent execution using Kung-Robinson Optimistic Concurrency Control (OCC), Tree-CRDT 3-way AST merge, and Linda tuple-space blackboard.


Related MCP server: onto-agent

🏛️ Tri-Pillar Cognitive Architecture

                                  ┌─────────────────────────────────────────┐
                                  │           SUPERVISOR / AGENT            │
                                  │  (Antigravity / OpenCode / Claude / AI) │
                                  └────────────────────┬────────────────────┘
                                                       │ JSON-RPC 2.0 (stdio)
                                                       ▼
┌───────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│                                            PROJECT CHRISTINA FAST-MCP GATEWAY                                     │
├──────────────────────────────────────┬────────────────────────────────────┬───────────────────────────────────────┤
│          PILLAR 1: RLM ENGINE        │       PILLAR 2: GRAPHIFY CPG       │        PILLAR 3: OKF MEMORY           │
│         (Context as Variable)        │          (AST Code Radar)          │         (Semantic Evolution)          │
├──────────────────────────────────────┼────────────────────────────────────┼───────────────────────────────────────┤
│ • Out-of-core Heap Variables         │ • In-Memory SQLite3 WAL Graph      │ • Sub-microsecond OKF Regex Parser    │
│ • Transactional CoW Rollback         │ • Sub-0.3ms Transitive Blast Query │ • Formal AGM Belief Revision          │
│ • 3-Phase EGRI Invariant Linter      │ • Composite God-Node Centrality    │ • Ebbinghaus Forgetting & GC          │
│ • Dynamic Governor (D<=2, N<=20)     │ • Circular Dependency & Wave Sort  │ • Symbol-to-ADR Constraints           │
└──────────────────────────────────────┴────────────────────────────────────┴───────────────────────────────────────┘
                                                       │
                                                       ▼
┌───────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│                                        DECENTRALIZED SWARM & DISTILLERY                                           │
├───────────────────────────────────────────────────────────────────┬───────────────────────────────────────────────┤
│                    SWARM CONCURRENCY BLACKBOARD                   │         GLOBAL EPISTEMIC DISTILLERY           │
├───────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
│ • Kung-Robinson Optimistic Concurrency Control (OCC)              │ • Zero-Information Leakage (I = 0)           │
│ • Semantic Tree-CRDT 3-Way AST Merge Resolution                   │ • Multi-Base Shannon Entropy Secret Redaction │
│ • Linda Tuple Space (`out`, `rd`, `in`, `eval`, `digest`)         │ • Bayesian Invariant Distillation (0700 Fence)│
└───────────────────────────────────────────────────────────────────┴───────────────────────────────────────────────┘

Mermaid Flowchart

flowchart TD
    Host[Host Agent Context] -->|JSON-RPC 2.0 stdio| MCP[Christina FastMCP Server]
    
    subgraph Trinity["Autonomous Cognitive Trinity"]
        MCP -->|cpg_query| CPG[Graphify CPG Radar<br/><i>sub-0.3ms SQLite WAL</i>]
        MCP -->|rlm_execute| RLM[RLM Sandbox<br/><i>Transactional CoW REPL</i>]
        MCP -->|memory_evolve| OKF[OKF Semantic Memory<br/><i>AGM Revision & Epistemic GC</i>]
    end
    
    subgraph SwarmDistill["Swarm & Cross-Project Substrate"]
        MCP -->|swarm_blackboard| OCC[Swarm Concurrency<br/><i>Kung-Robinson OCC + Tree-CRDT</i>]
        OKF -->|distill_global| Distill[Global Epistemic Distillery<br/><i>Anonymized Archetypes (I=0)</i>]
    end

    CPG -.->|Blast Radius & Waves| OCC
    RLM -.->|Isolated Sub-Calls| Host
    OKF -.->|ADR Anchor <=400 Tokens| Host

🚀 1-Command Zero-Friction Installation

Christina installs in seconds without background daemons, native compilation, or root privileges:

cd christina
./install.sh

Automated Diagnostic Preflight

./install.sh doctor
================================================================
 Christina Cognitive Amplifier Diagnostics (v1.0.0)
================================================================
✓ Python Version: 3.11+ (Requirement satisfied)
✓ SQLite3 in-memory WAL Engine: Ready
✓ OKF Semantic Memory Engine: Ready
✓ Google Antigravity Plugin: Registered
✓ OpenCode Plugin: Registered
✓ Global memory directory: ~/.config/christina/global_memory (0700 fenced)
================================================================
✅ All systems functional.

Run Full Test Suite (62/62 Passed in <1.2s)

./install.sh test
# Or with pytest directly:
pytest tests/ -v

🛠️ The Canonical FastMCP Tool Matrix

Christina exposes 4 unified JSON-RPC 2.0 stdio tools compatible with any MCP host:

Tool

Category

Key Capabilities & Modes

Performance Bound

cpg_query

AST Code Radar

blast_radius, def_use, cfg, vinterfaces, god_nodes, cycles, waves, info

$<0.3\text{ms}$ in-memory traversal

rlm_execute

Sandboxed Execution

Sub-agent MapReduce, transactional CoW REPL, 3-phase EGRI linting, compute governor

$D \le 2, N \le 20$ recursion bound

memory_evolve

Semantic Memory

search, persist, revise (AGM), gc (Ebbinghaus Mark-and-Sweep), distill_global, query_global

$6.72\ \mu\text{s}$/doc parsing

swarm_blackboard

Swarm Concurrency

out, rd, in, eval, occ_validate (Kung-Robinson OCC), tree_merge (Tree-CRDT), digest

Lock-free optimistic validation

MCP Tool Invocations

1. Blast Radius Analysis (cpg_query)

{
  "name": "cpg_query",
  "arguments": {
    "symbol": "PaymentGateway",
    "mode": "blast_radius",
    "depth": 3
  }
}

2. Transactional Sub-Query Execution (rlm_execute)

{
  "name": "rlm_execute",
  "arguments": {
    "query": "Audit error recovery handling in database connection pool",
    "scope_paths": ["src/db/", "src/pools/"]
  }
}

3. Semantic Memory Search with Symbol Constraint (memory_evolve)

{
  "name": "memory_evolve",
  "arguments": {
    "action": "search",
    "query": "zero-trust authentication policy",
    "linked_symbols": ["AuthMiddleware"]
  }
}

4. Swarm OCC Conflict Validation (swarm_blackboard)

{
  "name": "swarm_blackboard",
  "arguments": {
    "action": "occ_validate",
    "worker_id": "worker_security_audit",
    "read_version": 4,
    "read_set": ["AuthMiddleware", "SessionToken"],
    "write_set": ["AuthMiddleware"]
  }
}

⚡ CLI Quick Reference

# 1. Scan codebase and index AST knowledge graph into in-memory SQLite3 WAL
python3 -m christina.cli scan .

# 2. Calculate static blast radius for a target symbol
python3 -m christina.cli blast AuthService --depth 3

# 3. List top God Nodes ranked by composite degree centrality
python3 -m christina.cli god-nodes --limit 10

# 4. Search OKF institutional memory, ADRs, and symbol constraints
python3 -m christina.cli memory "connection pool timeout" --symbol DatabasePool

# 5. Run FastMCP JSON-RPC 2.0 stdio server
python3 -m christina.cli serve

# 6. Execute system diagnostics
python3 -m christina.cli doctor

🔌 Multi-Host Compatibility & Setup

Christina includes seamless 1-command registration across all standard AI agent hosts:

1. Google Antigravity CLI (agy)

Automatically symlinked to ~/.gemini/antigravity-cli/plugins/christina and registered in ~/.gemini/antigravity-cli/mcp_config.json.

2. OpenCode

Automatically registered in ~/.config/opencode/plugins/christina and ~/.config/opencode/opencode.json.

3. Claude Desktop

Automatically configured in ~/.claude/claude_desktop_config.json (or Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "christina": {
      "command": "python3",
      "args": ["-m", "christina.bridge.mcp_server"],
      "env": {
        "PYTHONPATH": "/path/to/christina"
      }
    }
  }
}

4. OpenAI Codex CLI

Automatically registered in ~/.config/codex/config.json.


🪮 Ponytail Ultra: Zero-Dependency Guarantee

Project Christina is strictly engineered under Ponytail Ultra discipline:

  • Zero External Runtime Dependencies: Uses Python standard library only (ast, sqlite3, re, json, hashlib, fcntl, math, pathlib, typing).

  • Zero Heavy Daemons: No Docker, no Neo4j, no ChromaDB, no Redis. In-memory SQLite3 WAL and POSIX file locks deliver sub-millisecond execution with $0\text{MB}$ idle background RAM.

  • Strict Code Brevity: Every module is concise ($<250$ LOC), cohesive, and self-contained.


📚 Documentation & Guides

Document

Description

Getting Started Guide

Step-by-step setup, verification, and first agent workflow.

Skills Taxonomy

Progressive disclosure skill definitions (cpg, rlm, memory, swarm).

Contributing Guide

Ponytail Ultra development guidelines, Nix setup, test suites, and ADR workflow.

Security Policy

Zero-information leakage ($\mathcal{I} = 0$), Shannon entropy secret scrubbing, and containment.

Changelog

Detailed release notes and version history.


📜 License

Project Christina is distributed under the open-source MIT License.

Copyright (c) 2026 Christina Engineering Swarm & Contributors.

Available Tools

4 tools
cpg_queryC

CPG & AST radar: def-use chains, CFG, virtual interfaces, blast radius.

Inputs: code, mode, depth, symbol.

memory_evolveC

Semantic memory: AGM revision, GC, distill_global, query_global, search, persist.

Inputs: code, tags, query, title, +6 more.

rlm_executeC

Sandboxed RLM REPL executor with transactional CoW rollback, 3-phase EGRI linter, governor.

Inputs: code, query, scope_paths.

swarm_blackboardD

Swarm blackboard & OCC concurrency: Kung-Robinson OCC validation, Tree-CRDT 3-way merge, Linda tuple space.

Inputs: action, channel, payload, read_set, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    C2.6/5.0

    Scored across 4 tools

    Disambiguation5/5

    Each tool targets a completely different subsystem: static code analysis, sandboxed execution, semantic memory, and concurrency/blackboard coordination. There is no overlap or plausible confusion between them.

    Naming Consistency4/5

    All tools follow a readable two-part snake_case format with domain-specific prefixes. However, the suffixes mix verbs (query, execute, evolve) with a noun (blackboard), creating a minor deviation from a fully uniform pattern.

    Tool Count5/5

    Four tools is a reasonable, focused count for a server whose scope appears to be a coordinated agent runtime: analyze, execute, remember, and coordinate. Each tool earns its place and none feel redundant.

    Completeness4/5

    The tool surface covers the major capabilities one would expect from this kind of system: code understanding, safe execution, memory persistence, and shared-state coordination. There are minor gaps such as explicit file/API interaction, but the broad sub-operations embedded in each tool make the overall surface feel fairly complete.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI agents to query and analyze code across multiple repositories through a unified knowledge graph, with tools for symbol search, impact analysis, and graph algorithms.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Provides AI coding agents with an embedded knowledge graph and architecture governance tools, enabling blast radius analysis, compliance checks, and smart context extraction from codebases.
      Apache 2.0
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables developer agents to perform semantic codebase search, dependency and impact analysis, cross-file refactoring, and full-stack API tracing through a unified query DSL over a high-performance graph engine.
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Provides persistent cross-session AI memory, code graph analysis, and subagent context compression so coding agents can resume work instantly and avoid re-reading the codebase.
      Elastic 2.0