CTX gives agents durable cognitive continuity across sessions, models, and tools.
CTX

Version 1.0.x

CTX

Cognitive Version Control for AI agents.

CTX turns long-running agent work into durable cognitive state: goals, tasks, hypotheses, evidence, decisions, conclusions, runbooks, and cognitive commits. Agents stop reconstructing context from chat and start from the current work line.

If CTX helps your agent workflow, star the repository so other builders can find it.

Minimum operator loop

Start from state, not from chat memory.

Run the helper, ask CTX what matters next, then request a focused planning packet before the agent touches code, docs, release flow, or recovery work.

ctx
ctx next
ctx plan --purpose "Plan the next work turn"

Install

The normal install path uses the published portable release. A local clone does not imply source mode unless you request it explicitly.

Windows

powershell -ExecutionPolicy Bypass -File .\install.ps1

Linux / macOS

bash ./install.sh
Installed tools

CTX exposes ctx, ctx-mcp, ctx-agent-acp, and the local Viewer from the install root.

Agent loop

CTX works best when agents treat it as the operating state for the task, not as an after-the-fact log. The packet returned by ctx plan includes the next work line, focused context, evidence, decisions, conclusions, and runbook suggestions.

1Inspect

ctx status, ctx audit, ctx next

2Plan

ctx plan returns focused context and playbooks.

3Work

Make the smallest useful change that produces evidence.

4Close

Record evidence, decisions, conclusions, and a cognitive commit.

MCP

The local MCP server lets Claude, Codex, VS Code, and other MCP-capable tools read CTX state through a controlled stdio adapter. Start read-only, smoke test ctx_plan, then enable write mode only when the agent should create artifacts.

{
  "mcpServers": {
    "ctx": {
      "type": "stdio",
      "command": "C:\\ctx\\bin\\ctx-mcp.cmd",
      "args": ["--repo", "C:\\path\\to\\ctx-repo", "--mode", "read-only"],
      "tools": ["*"]
    }
  }
}

Build your local MCP configuration

Viewer

CTX Viewer is the visual surface for cognitive history. It shows working context, commit timelines, graph lineage, details, origins, playbooks, and hypotheses without asking the agent to reload the whole repository into a prompt.

Working context keeps the active cognitive line visible.
Commit threads preserve evidence, decisions, and conclusions.

Cognitive model

CTX records reasoning as first-class artifacts. This makes agent work inspectable, auditable, resumable, and portable across sessions.

Goals Tasks Hypotheses Evidence Decisions Conclusions Runbooks Commits

Next steps