Roadmap

Sailfin's path from alpha to a stable 1.0 release and beyond.

Built July 20, 2026 from the Linear roadmap at deploy time.

Done In progress Planned

Release 1.0

0 of 36 items closed across 7 milestones

Accelerated ML

Enter the AI-compiler arena (Mojo/Triton/JAX/XLA) without rebuilding XLA: a shape-typed tensor IR reusing an existing accelerator substrate (Track A) plus capability-typed accelerators (Track B). Design: SFEP-0052.

0/1

ML Acceleration — Foundations

Product & Adoption

Adoption of Sailfin as a language agents write better than the alternatives. The MCP-firewall gateway is one demo of that thesis, not the thesis. Gated by SFN-350.

0/5

Test Infrastructure

A Sailfin-native, Vitest/cargo-test-grade workflow via the sfn/test capsule, plus parallel test execution (CI sharding + per-test cache) to keep the pre-1.0 loop fast

0/2

Build & Toolchain

Bring the self-hosted build to a Rust/Go/Zig-class experience: per-module memory budgets, faster cold builds, a slim Makefile with sfn owning build/check/bench, compiler decomposition into sub-capsules, CLI modularization, and native Windows self-host

0/10

Structured Concurrency

Mature the concurrency pillar from its working v0 (routines, channels, spawn/await, thread-pool scheduler) to a complete model: cancel-on-fault nurseries, an epoll/kqueue async-I/O reactor, and a bridged websocket runtime.

0/5

Effect System & Language Core

Harden the effects pillar and the language substrate beneath it: hierarchical effects, effect polymorphism, first-class function values, and capsule-defined decorators — the type-system keeping effect annotations honest across generics & higher-order code

0/4

Capability-Sealed Runtime

The 1.0 hallmark: extend Sailfin’s compile-time capability enforcement into a runtime seal, so a program provably cannot exceed its declared effects at run time. Spans the seal-sufficient native backend, an owned syscall layer, per-task capability context

0/9

Post-1.0

These workstreams begin after a stable self-hosted toolchain ships.

Enterprise Hardening & AI Integration

Library-level AI support, taint tracking, and advanced safety features — delivered as capsules, not language syntax.

5 items

sfn/ai capsule — model invocation, prompt templating, generation provenance (replaces the removed model/prompt/tool/pipeline keywords)
Secret<T> / PII<T> taint tracking integrated with the effect system
sfn lsp — language server for IDE integration
sfn audit enhancements — transitive capability analysis, SBOM generation
User-facing ownership & full borrow checking (&T, &mut T, shared borrows, lifetimes) — Phase U, widening the 1.0 runtime memory-safety floor (epic #1209) to user code.

Platform & Ecosystem

Runtime, tooling, and ecosystem infrastructure for post-1.0 growth.

6 items

Async runtime — Sailfin-native event loop and async I/O (the v0 task scheduler, channels, spawn/await, and parallel shipped pre-1.0; post-1.0 scope is the non-blocking event loop and TLS/keep-alive network integration)
Runtime diagnostics — structured tracing and allocation telemetry
WebAssembly emission target
Native test framework — golden, adversarial, and replay test types
|> pipeline operator with async and lazy semantics
GPU effect enforcement and tensor dispatch

Backlog & Exploration

Workstreams without an architected Linear initiative yet. Some will graduate into initiatives once design work lands; others stay exploratory.

Perf-parity native backend — the post-1.0 long tail beyond the seal-sufficient backend (#1640): full optimization and codegen maturity after the 1.0 capability seal lands
Structured concurrency (v0 shipped) — routine, channel, spawn/await, parallel, scheduler work today; atomics beyond channel-send and richer sync primitives are post-1.0
Effect system hardening — validate_effects() is a build gate (shipped); Raw-degradation holes closed for cast (#1627), ternary (#1690), and is type-guard (#1753); remaining: prefix */& and assignment-as-expression (#1180-c); hierarchical sub-effects (detection + manifest tightening) shipped (SFEP-0017); effect polymorphism post-1.0
Phase 2 containers — closures with capture, generic constraints, drop emission
Documentation expansion — production-quality site coverage, remove legacy stage references
LLM adoption levers — llms.txt, sfn check --json, MCP server, Rosetta Code corpus
Release pipeline hardening — signed checksums, installer CI, self-hosted-only artifacts
unsafe capability enforcement
Currency literals ($0.05) and time literals (1s, 150ms)
Notebook and interactive tooling
Effect handlers (algebraic effect semantics)

Contributing

Open an issue or start a discussion in a pull request. Tag it with the appropriate milestone label so it can be triaged against the 1.0 scope. Major design changes that affect the language or ABI should reference a written proposal in docs/proposals/ before implementation begins.

Open an Issue Contributor Guide