Context Engine

Every event across your org — Stripe, Slack, GitHub, Jira, LiveAgent, your site — streams into the Context Engine. It ranks, prunes, dedupes, and caches so every agent call carries only what matters.

The Context Engine is the brain of your agent. Sources stream events into a continuously updated index; when an agent asks a question, the engine scores the candidates, prunes the long tail, and serves a compact context pack — on average −89% tokens per query compared to passing the raw working set.

Create your grid#

Sign up at meaninggrid.harnext.dev to create a grid, then connect sources from the dashboard — Slack channels, git repositories, Stripe events, Jira projects, LiveAgent queues, website pages. Each source syncs continuously; there is no manual re-indexing.

Connect harnext#

Inside a repo, harnext setup links the harness to your grid. From then on every run builds its context through the engine automatically.

cd my-repo
harnext setup
# ✓ context engine connected · harness ready

Connect any other agent#

Add the engine as an MCP server and any MCP-capable agent can query the same grid:

{
  "mcpServers": {
    "context-engine": {
      "url": "https://meaninggrid.harnext.dev/mcp",
      "headers": { "Authorization": "Bearer $MEANINGGRID_API_KEY" }
    }
  }
}

What the engine does#

• Continuous index — sources stream events into the grid, so whatever your agent queries is always current.
• Relevance ranking — scores and orders candidates per query, then prunes the long tail before it ever reaches the model.
• Dedup & cache— collapses repeats and caches hot context, so repeat calls don't re-pay for the same tokens.
• Drop-in for any harness — one small HTTP + MCP surface for harnext, Claude Code, Codex, or your own loop.

Where to go next#