Skip to main content
These frequently asked questions cover what Contexity is, how it fits into your existing AI coding workflow, and how it handles trust, metrics, and portability.
Not in the generic sense. Contexity is a local-first context continuity layer for AI coding agents. It stores structured project context with provenance, freshness, run events, and metrics — not arbitrary text blobs. Every piece of context carries metadata about where it came from, when it was observed, and how confident the system is in it.
No. Contexity gives your agent better context and safer continuity. You still interact with Codex, Claude Code, Cursor, or another supported host — Contexity works in the background, supplying the agent with structured project knowledge through MCP so it can give you more accurate and consistent answers.
No. Normal users interact with their agent host. The host calls Contexity automatically through MCP. You only run CLI commands for initial setup, health checks, and explicit context management tasks such as invalidating a stale item or exporting a capsule.
No. Agent writes normally become candidate context unless a safer verification path applies. Candidate context is labeled as unverified and suppressed from verified-only retrieval, so it cannot silently pollute the knowledge base that other queries depend on.
They should be captured as external references with extracted signals. The Slack message or thread is the source of record. The signal is the actionable requirement, decision, warning, or research takeaway pulled from that message. Raw Slack text is never injected directly as model instructions — only the structured signals derived from it are surfaced to the agent.
Yes. Run the following command to disable visible metrics for a project:
contexity metrics disable --project /path/to/project
Structured metrics remain available for inspection at any time — disabling visibility does not delete the underlying data.
Not by themselves. Heuristic metrics are useful for internal diagnosis and product feedback, but marketing-grade proof requires paired A/B benchmarks with a Contexity-assisted run versus a baseline run, multiple valid trials, and confidence intervals. Use heuristics to guide your own understanding, not as standalone external claims.
No. Contexity is local-first. All project state lives in .cxcap/ and your local Contexity state home. Host setup, MCP serving, and context storage all operate from your local machine. No data is sent to an external service unless you explicitly configure an integration that does so.
Yes, using export and import capsules. A capsule lets you snapshot the context for a project and restore it on another machine or after switching branches. See the Import and Export guide for step-by-step instructions.