Wagle is a multi-project, multi-environment feature-flag and typed remote-config service where AI agents are first-class principals: 15 MCP tools to create, target, flip, and verify flags, every write audited under the agent's own identity, on the same trail as yours. Humans work the same flags from the console. Evaluation stays in your process: in memory, no database on the hot path.
15 mcp tools · revocable principals · evals localWhy & how it works, including where it's less strong →15 tools on /mcp: full flag and segment CRUD, state flips, targeting rules and % rollouts, history, and evaluate_flag returning the exact value and reason an SDK would compute. Agent-only, enforced server-side: a user token on the MCP path gets permission_denied.
An agent holds one static bearer; wagle exchanges it server-side for short 24h tokens, so no long-lived access token rides the wire. Revoke the agent and future exchanges fail while its live token dies on the blocklist. Every write bumps a per-project version and lands in history under the agent's own identity, never a shared key.
Every flag lives in an in-memory, (project, environment)-partitioned index. SDKs pull one snapshot and evaluate locally: a hash and a table lookup, no round-trip, ever. Server-only flags never leave the server.
xxHash64 over {project}:{env}:{flag}:{entity}, mod 10000. The same entity always lands in the same bucket, so a 5% rollout is a stable 5%.
Compose rules over segments and attributes with an ordered fall-through to a default variation. Typed values, not just booleans.
Flags are keyed by (project, environment) day one. Same flag set across dev, staging, prod, with a different rollout in each, promoted explicitly.
Untouched screenshots of the console working a real project: one bool flag, one string flag with a rolling split, one JSON config with a rule and a rollout, and the history those edits leave behind.
A sync is the single call that ever reaches wagle: your SDK asking for its (project, environment) config. Every evaluation runs locally in your process and is free, always. Nothing else is metered: not flags, not environments, not projects, not seats. Signing up puts you on Free with 500,000 syncs a month and no card.
evals free · flags unlimited · seats unlimited · no overageEvery plan is a capped bucket and there is no overage on any of them: outgrow one and we move you to the plan that fits. Your config keeps being served either way. SDK counts below are always-on instances at the SDK’s own 30-second default poll, about 87,600 syncs each per month.
Two inputs, one estimate. No usage archaeology at month end.
concurrent processes holding a snapshot: servers, pods, average concurrent browser sessions
how often each instance re-pulls (SDK default: 30s)
syncs / month = instances × (2,628,000 ÷ interval)
Evaluations never count: they run inside your process and are free, always. Only the config pull reaches wagle, so the estimate above is your whole billable volume.
What every account starts on: room for five always-on SDKs at the default poll.
Not a trial and not a subscription: it's what an account without one gets. No card, ever.
Side projects and small apps: a couple of dozen always-on SDKs.
No card at signup: subscribe when ready. Billed from day one, cancel anytime.
A real production fleet: around ninety always-on SDKs across your services.
No card at signup: subscribe when ready. Billed from day one, cancel anytime.
Several services and hundreds of pods, polling as fast as you like.
No card at signup: subscribe when ready. Billed from day one, cancel anytime.
For contractual guarantees: an uptime SLA, your own instance, your own database.