web-search Backends
Search backends are HTTP adapters in front of Exa, Firecrawl, Grok, or your own search infrastructure. Same contract whether the platform runs it or you do.
What counts as a backend
- Platform adapter — the platform ships small adapters for Exa, Firecrawl, and Grok out of the box. Admins register one platform backend per provider.
- Self-deployed — you host the adapter (inside your VPC, with your upstream keys) and register it for a single project. Pays only the platform circulation fee per search.
Ownership
- Platform — registered by an admin, visible to every project as a default option.
- Project — registered by a project owner, visible only to that project.
Namespace slugs
Each backend gets a stable namespace_id. You pick a slug; the platform composes the full id:
- Platform:
ctxkit.search.<slug> - Project:
proj.<project_id>.search.<slug>
Pick a slug that survives redeployments (include a version suffix like -v1). Re-registering with the same slug preserves billing continuity; changing the slug resets the cost cache.
Instances
A backend is a pool of instances — typically one HTTP process per upstream API key / region. The platform round-robins healthy instances and marks failing ones inactive.
Implementing your own
Your backend speaks the same HTTP + HMAC contract the platform backends use. The web-search project page has a detailed Implement tab with the full protocol, signing helpers, and a FastAPI scaffold.