Calmkeep

Why Calmkeep Exists

Most LLM usage is turn-based. Each response is strong on its own, but complex professional work rarely happens in isolation.

Backend systems evolve across dozens of prompts. Legal arguments compound across pages. Strategic decisions build on prior constraints.

Without structural continuity, models subtly reintroduce conflicting patterns, shift definitions, or drift from earlier architectural decisions.

Calmkeep reduces that drift. It preserves frameworks so reasoning compounds instead of resetting.

In testing, standard Claude introduced a validation pattern upgrade at turn 14 — then immediately abandoned it for the next three new modules, reverting to lower-rigor habits as if the upgrade never happened. Calmkeep held the pattern through turn 25 with zero backslide. That difference is the product.

The Core Difference

Standard LLM usage is episodic:

Each turn is processed independently
Earlier decisions are reinterpreted rather than preserved
Architectural patterns can subtly shift mid-session

Calmkeep introduces structured continuity:

Preserves earlier architectural decisions across long builds
Reduces reintroduction of conflicting libraries or patterns
Maintains definitional stability across extended analysis
Supports multi-step system evolution without silent drift
Enables compound reasoning over long conversational arcs

Drift isn't only architectural. In extended legal and strategic sessions, models under social or commercial pressure — a client who wants to close, a counterparty framing a concession as reasonable — tend to soften established positions without flagging that they've done so. Calmkeep holds the framework under that pressure too.

Where It Delivers Value

Long-session backend API development
Large schema evolution and refactor projects
Multi-tenant and RBAC-heavy system design
Legal and strategic reasoning that builds across documents

The 50th response reflects decisions from responses 1–49 — not just their text, but their structure.

Architecture

Your Application → Calmkeep Runtime → ANTHROPIC API (BYO Key) → Structured Response

Calmkeep does not modify model weights, alter training data, or introduce hidden memory. It operates as an external orchestration layer around your existing Claude usage.

Bring your own ANTHROPIC API key
Claude billing remains directly with Anthropic
No token resale or proxy billing
No safeguard bypassing
External runtime only

What You Receive

Calmkeep SaaS runtime endpoint
Python SDK
Runtime continuity layer
Input/output guardrails enforcement
Runtime performance instrumentation
Structured response telemetry

ANTHROPIC API usage is billed separately through your own Anthropic account.

from calmkeep import CalmkeepClient

client = CalmkeepClient(
    calmkeep_key="YOUR_CALMKEEP_SUBSCRIPTION_KEY",
    anthropic_key="YOUR_ANTHROPIC_API_KEY"
)

response = client.complete(
    prompt="Continue evolving the existing multi-tenant API..."
)

Accessing Calmkeep

There are three ways to use Calmkeep depending on your workflow.

1. Python SDK Integration

Install the Calmkeep SDK and route your ANTHROPIC API requests through the Calmkeep runtime continuity layer.

pip install git+https://github.com/calmkeepai-cloud/calmkeep_sdk

Initialize the client using your Calmkeep subscription key and ANTHROPIC API key.

from calmkeep import CalmkeepClient

client = CalmkeepClient(
    calmkeep_key="YOUR_CALMKEEP_SUBSCRIPTION_KEY",
    anthropic_key="YOUR_ANTHROPIC_API_KEY"
)

2. MCP Connector (Claude Desktop / Cursor)

Developers using Model Context Protocol (MCP) clients such as Claude Desktop or Cursor can route requests through the Calmkeep runtime using the official Calmkeep MCP connector.

The MCP server acts as a lightweight bridge that forwards prompts to Calmkeep while preserving the runtime continuity layer that stabilizes long multi-step reasoning workflows. In this format the Claude App is calling Calmkeep, meaning the Claude application calls Calmkeep as a MCP tool. This preserves frameworks and increases constraint across multi-step workflows, though it differs from the full session experience available at calmkeep.ai/chat.

git clone https://github.com/calmkeepai-cloud/calmkeep-mcp
cd calmkeep-mcp
pip install -r requirements.txt

# configure environment variables
export CALMKEEP_API_KEY=your_calmkeep_subscription_key
export ANTHROPIC_API_KEY=your_ANTHROPIC_key

python mcp_server.py

After launching the MCP server, the calmkeep_chat tool becomes available to compatible MCP clients.

MCP repository: github.com/calmkeepai-cloud/calmkeep-mcp

3. Claude Code Plugin

Developers using Claude Code can install the Calmkeep plugin directly:

/plugin marketplace add calmkeepai-cloud/calmkeep-claude-plugin
/plugin install calmkeep-continuity

After installation, configure your API keys:

export CALMKEEP_API_KEY=YOUR_CALMKEEP_SUBSCRIPTION_KEY
export ANTHROPIC_API_KEY=YOUR_ANTHROPIC_API_KEY

The plugin routes Claude Code requests through the Calmkeep runtime, allowing long multi-step development workflows to maintain architectural continuity.

Plugin repository: github.com/calmkeepai-cloud/calmkeep-claude-plugin

Professional Plan

About Calmkeep