Kontexta
Context management tool with agent inter-operability and fully local.
Documentation
Kontexta: workspace + context engine for coding agentsREV 6.1.0
Your agents have memory. It just isn't yours.
Every coding agent keeps its own silo. kontexta is one local workspace they all share: searchable, versioned, portable.
$ npx kontexta start
FIG. 0: LIVE LOOP
Claude Code fixes the bug and journals it. Hermes, a different tool with zero prior context, recalls it from the same local vault. Wire up your clients →
66.8%
Fewer file tokens loaded per task
3.0×
Less context to reach the same files
41,171
Tokens saved across 4 scenarios
OPERATING PRINCIPLE
kontexta turns AI agents from isolated chat sessions into a continuous operational system.
npx
Recommended
One command. Dashboard + MCP together. Requires Node 22 LTS.
View procedure →✓ Dashboard + MCP
Docker
Alternative
Persistent vault on a host volume.
View procedure →✓ Dashboard + MCP
MCP only
Headless
Headless MCP server from npm. No dashboard, no ports. Requires Node 22 LTS.
View procedure →MCP only
Agent Skill
No MCP
For agents whose client blocks MCP servers (managed/enterprise installs). Python bridge the agent runs from its shell.
View procedure →Shell bridge
Contributors
Advanced
Clone, bootstrap, hack. Apache-2.0.
View procedure →✓ Dashboard + MCP
AGENT INTEGRATION: MCP INSTALLER & AUTO-JOURNALING
Connect MCP & Enable Coding Agents
Kontexta connects to your AI coding agents in seconds. Use our automated MCP installer to register kxta directly into your agent configs with zero JSON editing, and wire lightweight hooks that stream prompts, replies, and shell commands into your journal.
Claude CodeCursorGemini CLICodexCopilotClineContinueHermesAntigravityWindsurf
1-Click MCP InstallZero JSON edit
$ npx kontexta connect --agent claude-code,cursor
Registers the kxta MCP server into your agent's native config file, or run npx kontexta connect to detect all agents.
Enable Journal HooksMulti-agent list
$ npx kontexta hooks enable --agent claude-code,gemini
Enables agents and configures their native hooks to capture prompts, outputs, and shell actions.
Inspect & WebUIVerified status
$ npx kontexta hooks status
Or connect MCP and toggle hooks with a single click in the WebUI via Configure → AGENTS or during setup.
[
Explore all 58 tools →
The full MCP inventory: search, surgical reads, Hands, reports, journaling, calendar.
](https://kontexta.dev/tools)[
Publish your docs →
Vault → documentation sites, API references, and llms.txt.
](https://kontexta.dev/publish)[
How it compares →
An honest table vs CLAUDE.md, vendor memory, mem0, Linksee, and Cosmos.
](https://kontexta.dev/compare)[
Set up your client →
Config snippets for all 10 supported agents.
Where does my data live?
On your machine. By default, data lands at ~/.local/share/kontexta (Linux), ~/Library/Application Support/kontexta (macOS), or %APPDATA%\kontexta (Windows), no configuration needed. You can override this with KONTEXTA_DATA_DIR. Everything is indexed in a local SQLite database. Nothing is sent to a third-party service: no embeddings API, no telemetry, no cloud sync. Your only outbound traffic is the optional git push to a remote you configure yourself.
Does it work offline?
Fully. The MCP server, the web UI, and the SQLite index are all local. Git push for backup is optional and the only operation that touches the network.
Which agents work with kontexta?
Anything that speaks MCP: Claude Code, Cursor, Gemini CLI, Codex, Continue, Antigravity, Claude Desktop, Aider, Cline, and OpenClaw. The admin_onboard_agent tool auto-detects and writes the right context file (CLAUDE.md, AGENTS.md, GEMINI.md,.cursor/rules/*,.continue/rules/*,.aider/kontexta.md,.cline/kontexta.md,.openclaw/kontexta.md) so agents pick up the workflow automatically.
How is Hands execution sandboxed?
Hands tools are declared per-project in kontexta.json, so agents cannot invent commands. Each invocation runs with no shell interpreter, a locked working directory, and a scrubbed environment (only PATH, HOME, USER, LANG, TZ are forwarded). Output is capped via a ring buffer; timeouts kill the entire process group. Argv parameters are validated by re2 patterns (ReDoS-proof) with an 8 KB UTF-8 cap, and the default pattern blocks leading-dash argv injection.
Can I require human approval before destructive commands run?
Yes. Mark a Hands tool as confirm-required and it returns a single-use, CSPRNG-bound token that expires in 60 seconds. The agent must call hands_confirm with that token to actually execute. The token is the auth, and it cannot be replayed.
What's the license?
Apache-2.0. Use it commercially, fork it, embed it. No telemetry callback, no rug pull.