f69
Feature flags evaluated at the edge: a Rust -> WASM evaluator running on edge runtimes against a versioned manifest: deterministic, fail-closed, and no round-trip to a vendor.
The problem
Most flag platforms evaluate centrally: your app calls a vendor endpoint with the user context, the vendor decides true/false, then you unblock the render. That works until the vendor is far away, slow, or down, and ~100 ms per decision is exactly the latency that causes UI flicker. f69 inverts the model: the manifest lives next to your code, and a small deterministic evaluator makes the call locally.
Why edge evaluation
A Rust -> WASM evaluator runs on your edge runtime against a versioned manifest synced globally. No network call in the render path: the same evaluator returns the same answer everywhere, byte-for-byte, in your edge worker, your server, or a CLI.
The three promises
No flicker: decisions are local, nothing to wait on. Predictable rollouts: a frozen bucket hash means a 10% rollout picks the same 10% on every runtime. Fail-closed by default: a malformed flag falls back to its declared default; one bad flag can’t break the rest.
The planes
Control plane
The dashboard: author flags, manage environments, mint service-account tokens. Everything that changes workspace state happens here.
Logic plane
A versioned manifest (rules, rollouts, segments) synced globally to every runtime.
Edge plane
The WASM evaluator and the edge API your app calls every request.
The edge contract
| Endpoint | Purpose | Auth |
|---|---|---|
POST /v1/identify |
register a user/entity for a workspace | service-account bearer token |
POST /v1/evaluate |
evaluate flags for that entity | same token, scoped to one environment |
The same Rust evaluator is also available as a library (and as WebAssembly), so SDKs can resolve flags fully offline with byte-for-byte identical results. The offline and online paths physically cannot disagree: they’re the same code.
Determinism & safety
Two properties do most of the work:
- Frozen bucket hashing. Rollout math is deterministic across runtimes, so no coordination is needed between edge nodes to agree on who’s in a rollout.
- Fail-closed evaluation. Malformed flags degrade to their declared default instead of throwing. The safety path is the default path.
A security posture, not a footnote
The edge surface is read-only and authenticated by environment-scoped service accounts; the dashboard supports passkeys for teams that need them. The runtime surface stays tiny and auditable: the edge never mutates state.
As a product
Billing is metered on the axes that match actual usage: monthly tracked entities (MTEs), evaluations, and service accounts.
| Free | Standard | |
|---|---|---|
| Price | $0 / month | $40 / month base |
| Tracked entities | 5,000 | 100,000, then usage-based |
| Evaluations | 1M / month | 10M / month, then usage-based |
| Environments | 2 | Unlimited |
| Service accounts | 2 | 10 |
| Targeting rules | Basic predicates | Advanced (%, rollouts, schedules) |
Self-hosting is a documented path for teams that can’t send flag state to a third party.