[Meta] Improvement program 2026 — comparison, roadmap & tracking

[pathosDev]

Tracking issue for the 2026 improvement program: framework comparison → gap analysis (deduplicated against the existing tracker) → roadmap. 37 new issues were opened as part of this program (#406–#446); everything else referenced below already existed.

1. Where actor-ts stands (framework comparison)

Capability	Akka 3 / Pekko	Orleans	Erlang/Elixir	Proto.Actor	Temporal/Restate	Cloudflare DO	actor-ts
Core actors + supervision	✅	✅ (grains)	✅	✅	—	partial	✅
Cluster + sharding + singleton	✅	✅	✅	✅	—	managed	✅
Event sourcing + projections	✅	partial	Mnesia	basic	✅ (workflows)	—	✅ (5 backends)
CRDTs / distributed data	✅	—	—	—	—	—	✅ (8 types)
Reliable delivery	✅	✅	—	—	✅	—	✅
Streams w/ backpressure	✅	Orleans Streams	GenStage/Broadway	—	—	—	❌ → #147/#54
Durable timers/reminders	—	✅	—	—	✅	✅ (alarms)	❌ → #168
Saga / durable workflows	✅ (Akka 3)	—	—	—	✅	—	❌ → #179/#88
AI-agent runtime	✅ (2025 pivot)	via MS AF	—	—	✅ (headline)	—	❌ → #421 (flagship)
DevTools / observer UI	commercial	dashboard	✅ observer	—	✅ UI	—	❌ → #445
Browser/edge story	—	—	—	—	—	✅	❌ → #446 (unique opportunity)
Multi-runtime TS-native	—	—	—	—	SDK only	Workers only	✅ Bun+Node+Deno

Verdict: ~95 % parity with the classic Akka core; the open gaps were almost all already ticketed here. The differentiators nobody else can copy cheaply: TS-native full stack on three runtimes, and (upcoming) durable AI agents + isomorphic actors.

2. Positioning

"The TypeScript-native actor runtime — and the durable AI-agent runtime for TS." Batteries included, dependencies excluded.

3. Dependency policy (binding — details in #419)

• Zero-runtime-dependency core as the goal (after [Feature] Built-in zero-dependency HTTP backend as default; Fastify becomes an optional peer #413 fastify → optional peer; [Feature] Dependency policy: zero-runtime-dependency core (evaluate inlining ts-pattern) #419 decides ts-pattern, which gates [Feature] Replace the ClusterRouter cluster-event instanceof pair with match() #235–[Feature] Replace RefCodec instanceof-chains with match() #247).
• Integrations only as optional, lazily imported peerDependencies (BrokerActor pattern).
• Prefer SDK-free implementations over the built-in HttpClient where reasonable (Ollama/Anthropic/OpenAI HTTP, MCP JSON-RPC, etcd/Consul, libSQL/D1). devDependencies are unrestricted.
• DevTools UI ships bundled/vanilla — no UI framework as a runtime dependency.

4. Documentation definition-of-done (binding — details in #420)

No feature is done without: concept guide (EN + DE), full JSDoc on public APIs, runnable example, HOCON config reference, CHANGELOG entry. Large modules (agents, jobs, client, devtools) additionally get their own docs chapter + example app.

5. Milestones (order, not version promises — versions are assigned at release cut)

M1 — Security hardening (first)

Foundation (land early, everything after inherits it):

• [Feature] Adopt Biome as the repo formatter + linter #417 Biome formatter + linter
• [Test] Shared awaitCondition helper to replace sleep-based waits #418 awaitCondition test helper
• [Feature] Dependency policy: zero-runtime-dependency core (evaluate inlining ts-pattern) #419 dependency policy + ts-pattern decision (gates [Feature] Replace the ClusterRouter cluster-event instanceof pair with match() #235–[Feature] Replace RefCodec instanceof-chains with match() #247 and [Feature] Built-in zero-dependency HTTP backend as default; Fastify becomes an optional peer #413 design)
• [Docs] Documentation Definition-of-Done for feature work (PR template + checklist) #420 documentation DoD (PR template)

Security triage of the open backlog, in waves — most urgent first: #110, #121, #126 + #136, #117 + #141, #406.

• S1 input validation & injection: [Security] Tag strings may flow unescaped into CQL/SQL queries #136, [Security] PersistenceId accepts arbitrary characters (NUL, .., control chars) #133, [Security] Actor names accept control chars, allowing log injection #134, [Security] ActorPath.child(name) doesn't reject path-traversal segments #126, [Security] Object-storage list() returns malformed keys not validated #123, [Security] HTTP redirect target not validated (open redirect risk) #125, fix(io/tcp): TcpSocketOptionsValidator should validate the nested framing.maxLineLen #372, feat(io/sse): make the SSE inbound buffer cap an SseOptions field #371, [Security] Guard HOCON deepMerge and key-path assignment against prototype pollution #406 (new: HOCON prototype pollution) — [Security] Guard HOCON deepMerge and key-path assignment against prototype pollution #406 done (three further vectors found while fixing: single-segment __proto__, ${__proto__} substitution reads, and deepMerge carrying an own __proto__)
• S2 crypto correctness: [Security] AES-GCM IV reuse not prevented by wrapper API #110, [Security] Key-version byte overflow after >256 rotations #111, [Security] HKDF info-parameter is static across deployments #108, [Security] No default snapshot-policy guidance for DurableState #107
• S3 cluster protocol auth: [Security] ClusterClient envelope.from spoofing reroutes replies #121, [Security] Joining-member version-hijack via pre-claimed higher version #114, [Security] Gossip-replay attack via missing timestamp validation #112, [Security] ClusterClient TLS config not validated at construction time #144, [Security] SeedProvider DNS resolution not pinned, vulnerable to DNS hijack #145, [Security] Heartbeat sequence-number accepts NaN / Infinity #115
• S4 unbounded growth / DoS: [Security] Receptionist subscribers set is unbounded #137, [Security] Cluster member map grows unbounded under adversarial gossip #138, [Security] DistributedPubSubMediator per-topic subscriber list unbounded #139, [Security] Tombstone TTL never expires when removedAt is Infinity / garbage #113, [Security] Global ask-counter overflow after 2^31 calls #119, [Security] Prometheus cardinality attack via user-controlled label values #131, [Security] JSON.stringify on error paths can hang on circular references #146, [Bug] BoundedMailbox drop-head overflow is a no-op while the mailbox is suspended #407 (new: BoundedMailbox suspended drop-head bug) — [Bug] BoundedMailbox drop-head overflow is a no-op while the mailbox is suspended #407 done; note the suspension window is ActorCell.failToParent, i.e. every supervision escalation, so the default bounded mailbox lost its bound exactly when it mattered
• S5 HTTP headers & info-leak: [Security] Missing X-Content-Type-Options: nosniff header on responses #127, [Security] CORS default may be too permissive #128, [Security] MDC context leak across async-storage tenant boundaries #129, [Security] Traceparent values written to plain-text debug logs #132, [Security] Error messages leak filesystem paths / stack frames in user-facing responses #130, feat(http): built-in dependency-free HTML sanitizer for untrusted markup #355, fix(http): default 500 response leaks err.message across all three backends #356, fix(http): request body-size limit parity across backends #357, feat(http/ws): per-route / configurable WS transport frame cap on Express + Fastify #373
• S6 races & resource leaks: [Security] DurableState ETag-refresh race in upsert path #117, [Security] setIfAbsent atomicity is backend-dependent under high contention #141, [Security] PersistentActor accepts out-of-order events from snapshot store #122, [Security] PersistentFSM stateTimeout race on rapid transitions #143, [Security] DistributedData pending-writes map can accumulate stale timeouts #140, [Security] PromiseActorRef.tell timer leak on duplicate reply #118, [Security] SqliteJournal lock contention via filesystem flock() #124
• S7 persistence robustness (in progress in a parallel worktree features/persistence-hardening — tracked here, not re-planned): [Bug] Silent in-memory fallback masks a misconfigured persistence plugin (data-loss footgun) #377, [Bug] Injected shared pool ended by individual store close() (Postgres/MariaDB); object-storage backend double-close #378, [Bug] Journal highestSeq rewinds to 0 after a full delete (relational + in-memory) #379, [Bug] Cassandra journal: start() not single-flight; consistency option dead; misleading logged-batch comment #380, [Feature] Remove dead persistence API surface (batchSize/clock, durableStatePluginId, recovery.mode, consistency) #381, [Bug] Broken persistence examples (scylla-ledger snapshotPolicy, cassandra-plugin-hello import path) #382, [Bug] Inconsistent JournalError wrapping + doubled 'require requires' hint + non-uniform lazy-import #383

M2 — Performance & robustness

• [Feature] Ring-buffer backing store for Mailbox to remove O(n) shift() #408 ring-buffer mailbox (O(1) dequeue)
• [Feature] Per-actor message batching (throughput) in ActorCell.run #409 per-actor message batching (target: tell/ask ≥ 2×)
• [Bug] Route dispatcher errors through the system Logger/EventStream #410 dispatcher errors → Logger/EventStream
• [Feature] Reduce per-message allocations on the ActorCell hot path #411 ActorCell hot-path allocation reduction
• [Feature] Memoize ActorPath.toString() #412 memoize ActorPath.toString() (scheduled in the current release window)
• bench baseline committed + non-blocking CI bench job (advances Performance benchmarks vs Akka.NET / Akka (JVM) #27)

M3 — Packaging, DX & CI

• [Feature] Built-in zero-dependency HTTP backend as default; Fastify becomes an optional peer #413 native zero-dependency HTTP backend; fastify → optional peer
• [Feature] Add subpath exports to package.json #414 subpath exports
• [Feature] Add "sideEffects": false for tree-shaking #415 sideEffects: false (scheduled in the current release window)
• [Feature] Add publint, arethetypeswrong, and knip to CI #416 publint + arethetypeswrong + knip in CI (scheduled in the current release window; [Feature] Add subpath exports to package.json #414 sequenced after it, since its acceptance criteria depend on these tools existing)
• [Feature] Publish to JSR for first-class Deno consumption #442 JSR publish for Deno
• extend [CI] ParallelMultiNodeSpec + LeaseMajority quarantined on hosted runners (Bun worker respawn dies after first test) #325: Dockerized CI job for the quarantined worker-thread suites, then ratchet the coverage floor 80 → 89 (relates [Test] Parallel-test flake catalog, stress verification and diagnosis doc #290, Tests: S3ObjectStorageBackend integration via LocalStack #297, [Feature] Type-assertion debt audit (as unknown as T / as any) #256)

M4 — Flagship: actor-ts/agents

• [Feature] actor-ts/agents — LLM agents as supervised, durable actors (epic) #421 epic — MVP: [Feature] agents: provider-neutral ChatModel interface + SDK-free OpenAI / Anthropic / Ollama adapters #422, [Feature] agents: ToolRegistry with tools-as-actors (absorbs #202) #423, [Feature] agents: AgentActor — event-sourced conversation loop #424, [Feature] agents: deterministic testkit — Scripted / Recording / Replay ChatModel #425, [Feature] agents: OTel GenAI semconv spans + Prometheus metrics #426 — follow-ups: [Feature] agents: MCP client — consume external MCP servers as tools (SDK-free) #427, [Feature] agents: MCP server — expose actor tools over MCP (SDK-free) #428, [Feature] agents: orchestration — agent-as-tool, handoff, sharded agents, typed facade #429, [Feature] agents: human-in-the-loop approval gate #430
• [Feature] LLM-Tool-Call-as-Actor primitive #202 is absorbed by [Feature] actor-ts/agents — LLM agents as supervised, durable actors (epic) #421 — closed as absorbed, after verifying each part is carried: [Feature] agents: ToolRegistry with tools-as-actors (absorbs #202) #423 (tool dispatch/retry/circuit-breaking), [Feature] agents: AgentActor — event-sourced conversation loop #424 (durable per-conversation state, crash recovery, token/cost audit), [Feature] agents: OTel GenAI semconv spans + Prometheus metrics #426 (OTel + Prometheus), [Feature] agents: orchestration — agent-as-tool, handoff, sharded agents, typed facade #429 (cluster scaling)

M5 — Feature parity & patterns

Existing: #147 (streams subset first; #54 full DSL after), #168 (reminders), #179 + #88 (saga), #169 (placement), #170 (stateless workers), #156, #151, #159, #177.
New: #431 (jobs — after #168), #433 (DLQ), #434 (OpenAPI), #432 (presence).

M6 — DevTools & simulation

• [Meta] DevTools suite — one product from visualizer, time-travel, tracing, profiling #445 DevTools umbrella → [Feature] Live cluster visualizer (shipped in package) #204, [Feature] Dev-mode tracing visualizer (Flame-graph of message flows) #217, [Feature] Per-actor explain-plan (last-N message timing breakdown) #218, [Feature] Time-travel debugger over journal #201, [Feature] In-process actor flame-graph profiler #226 — complete, closed; all five panels shipped on one versioned tap protocol (DEVTOOLS_PROTOCOL_VERSION = 1) behind a ./devtools subpath export
• [Feature] Deterministic Simulation Testing (FoundationDB-style) #200 deterministic simulation testing (recommend raising priority — flagship-grade differentiator; [Feature] agents: deterministic testkit — Scripted / Recording / Replay ChatModel #425 contributes the agents leg)

M7 — Isomorphic actors (browser & edge)

• [Meta] Isomorphic actors — browser & edge program #446 program umbrella → [Feature] actor-ts/client — typed browser/frontend client (ask over WebSocket/HTTP) #435 (typed client), [Feature] React/Vue/Svelte bindings for the actor-ts client (useActor, useAsk, useTopic) #436 (bindings), [Feature] Browser-side ActorSystem with Web Workers #210, [Feature] Cross-runtime cluster (Bun + Node + Deno + Browser in one cluster) #215, [Feature] WASM/Edge-runtime subpackage (Cloudflare Workers / Deno Deploy) #209 (+ D1/libSQL backends from [Feature] Additional persistence backends: MongoDB, DynamoDB, MSSQL, libSQL/Turso, Cloudflare D1 (umbrella) #438)

M8 — Integrations (continuous, demand-driven)

New: #437 (cloud brokers), #438 (persistence backends) — DONE, closed: MongoDB, DynamoDB, MS SQL, libSQL/Turso and Cloudflare D1 all shipped with the full journal + snapshot + durable-state trio, EN+DE docs and live integration suites; follow-ons split out as #490 and #491, #439 (lease backends), #440 (observability pack), #441 (Helm), #443 (Effect interop — parked), #444 (delta CRDTs — parked).
Existing same category: #15, #16, #72, #73, #74, #160, #4, #5.

6. Deliberately NOT re-ticketed (dedup)

Streams (#54/#147), reminders (#168), saga (#179/#88), virtual-actor features (#169–#175), gen_statem/trap_exit/via-registry/RPC (#163–#166), multi-DC (#149), simulation/time-travel/hot-reload/chaos/visualizer (#200–#206), edge/browser (#209/#210/#215), Avro/Protobuf (#73), schema-registry client (#72), router strategies (#153/#154/#69), the code-quality catalog (#235–#282), CRDT property tests (done), CI coverage restore (#325), ZeroMQ/STOMP (#15/#16), HTML sanitizer (#355), persistence robustness (#377–#383, parallel session).

7. 1.0 criteria

M1–M3 complete; publint/attw/knip green; coverage floor restored to 89; bench gates holding; security backlog empty; API freeze + CHANGELOG discipline.

[pathosDev]

Tracker sweep + release-window progress.

Milestones completed

• M6 — [Meta] DevTools suite — one product from visualizer, time-travel, tracing, profiling #445 (DevTools) — closed. All five sub-issues done, seven panels on one versioned tap protocol, ./devtools subpath export, UI bundled so it works from a fresh clone.
• M8 — [Feature] Additional persistence backends: MongoDB, DynamoDB, MSSQL, libSQL/Turso, Cloudflare D1 (umbrella) #438 (persistence backends) — closed. Five backends, each with the full trio, EN+DE docs, live integration suites. Two follow-ons split out rather than left implicit: [Feature] CassandraDurableStateStore via lightweight transactions (LWT) #490 (CassandraDurableStateStore via LWT) and [Feature] Collapse SqliteJournal onto RelationalJournal via the new SqliteClient #491 (collapse SqliteJournal onto RelationalJournal).

Landed from the security backlog

• [Security] Guard HOCON deepMerge and key-path assignment against prototype pollution #406 (S1) — HOCON prototype pollution. Reproduced first: __proto__.polluted = true parsed to an innocuous {} while polluting Object.prototype process-wide. Three vectors beyond the issue turned up — a single-segment "__proto__" { … } replacing the config object's prototype, ${__proto__} reading through the chain, and deepMerge carrying an own __proto__ through ({ ...base } included).
• [Bug] BoundedMailbox drop-head overflow is a no-op while the mailbox is suspended #407 (S4) — BoundedMailbox drop-head. Worse than filed: the suspension window is ActorCell.failToParent, i.e. every supervision escalation, so the default bounded mailbox (10 000 / drop-head since v0.10.0) lost its bound precisely when a queue fills.

Also landed this window (audit bugs, outside the M-list)

• [Bug] TypedActor never delivers the terminated Signal to onSignal #448 — TypedActor never delivered the terminated signal. The kind was declared, documented and tabulated, but constructed nowhere; receiveWithSignal had zero test coverage, which is how a documented public feature stayed entirely non-functional.
• [Bug] Router never prunes terminated routees #449 — Router never pruned terminated routees: 1/N of round-robin traffic to dead letters, one lost message per Broadcast.
• [Bug] ProducerController skips confirm() for in-flight sends on shutdown #451 — ProducerController dropped in-flight confirmations on shutdown, so callers hung forever. With the default windowSize of 16 this was the common case, not an edge case.
• [Bug] FailureDetector downAfterMs semantics contradict its JSDoc/validator #452 — FailureDetector.downAfterMs documented as additive, implemented as absolute; downAfterMs <= unreachableAfterMs made the unreachable state dead code with nothing rejecting it.
• [Feature] Validate Router pool size > 0 (fail-fast) #455 — Router pool size <= 0 silently dead-lettered 100% of traffic.

One correction worth recording: the #449 analysis initially claimed the unconsumed Terminated was delivered to a live routee as work. It is not — ActorCell:799 only delivers a Terminated to an actor actually watching the subject. Caught because two regression tests passed without the fix; they were rebuilt against the dead-letter stream, where the loss is genuinely observable.

Tracker cleanup

17 issues closed with evidence, 4 filed for residual scope (#488–#491), 2 rescoped (#235 to the one real ClusterRouter site, #290 to the catalog + stress + doc that actually remain), and #54/#147 plus #69/#154 cross-referenced and kept open — #147 covers a deliberate ~5% subset, so closing #54 would have moved the full-DSL scope out of the tracker.

Section 6 of this issue can drop its saga (#179/#88) dedup note: #88 is now closed into #179 with its API sketch transferred.

[pathosDev]

Post-v0.12.0 planning round — 35 new issues

A full planning pass after the v0.12.0 release (repo state + module inventory + quality audit + adversarially verified competitive research on Rivet / Restate / Temporal / Akka / Pekko / Orleans / nact / Dapr). New issues, grouped:

Post-release cleanup

• [Docs] Refresh ROADMAP.md for the v0.12.0 release #527 [Docs] Refresh ROADMAP.md for v0.12.0 (also resolves the [Feature] ShardCommand types (StartEntity, GetShardStats, GetClusterShardingStats) #151 duplication)
• [Feature] Benchmark regression gate: stored baselines, nightly comparison, latency percentiles #528 [Feature] Benchmark regression gate (baselines, nightly comparison, percentiles)
• [Feature] Recover the v0.12.0 sharding throughput regression (-28 % single-node / -41 % two-node ask/s) #529 [Feature] Recover the v0.12.0 sharding throughput regression (-28 % / -41 %)

Features

• [Feature] actor-ts/workflows — durable execution on top of event sourcing (epic) #530 [Feature] actor-ts/workflows — durable execution on event sourcing (epic) — the flagship differentiator; every competitor (Rivet, Restate, Temporal, Akka Orchestration) is converging here
• [Feature] Virtual actor facade — keyed, auto-lifecycle entities on sharding + durable state #531 [Feature] Virtual actor facade (Orleans/Restate shape)
• [Feature] PersistenceQuery for the remaining backends (MsSQL, libSQL, D1, DynamoDB, object storage) + shared query base #532 [Feature] PersistenceQuery for the remaining backends + shared query base
• [Feature] Typed API: context.self, context.ask, messageAdapter, pipeToSelf + typed routers #533 [Feature] Typed API: context.self/ask, messageAdapter, pipeToSelf + typed routers
• [Feature] Typed EventSourcedBehavior / DurableStateBehavior #534 [Feature] Typed EventSourcedBehavior / DurableStateBehavior
• [Feature] ReplicatedEventSourcedActor: vector-clock GC + snapshot support #535 [Feature] ReplicatedEventSourcedActor: vector-clock GC + snapshots
• [Feature] Public persistence testkit: journal/snapshot conformance suite + failure injection #536 [Feature] Public persistence testkit (conformance + failure injection)
• [Feature] HOCON completion: include directives, concatenation, triple-quoted strings #537 [Feature] HOCON completion (includes, concatenation, triple-quoted strings; fixes [Security] HOCON parser silently ignores include directives #135)

Quality

• [Test] Compile and smoke-run examples/ in CI #545 [Test] Compile and smoke-run examples/ in CI
• [Test] Nightly workflow for the ACTOR_TS_SKIP_FLAKY_MNS-quarantined suites #538 [Test] Nightly workflow for the quarantined flaky suites
• [Feature] Security scanning: CodeQL, dependency review/OSV, SECURITY.md #539 [Feature] Security scanning: CodeQL, dependency review/OSV, SECURITY.md
• [Test] Make typecheck:dev green and gate tests/ + examples/ type-checking in CI #540 [Test] Make typecheck:dev green + gate it
• [Test] Coverage gate: single parser, ratcheted floor, per-module floors #541 [Test] Coverage gate: single parser, ratcheted floor
• [Test] Close unit-test gaps: devtools/cluster federation, typed/, runtime/tcp, delivery, io #542 [Test] Close unit-test gaps (devtools federation, typed, runtime/tcp, delivery, io)
• [Test] Expand property-based testing (fast-check): serialization, HOCON, sharding, migrations, mailboxes #543 [Test] Expand property-based testing
• [Docs] Docs CI: build-only check on PRs, link checker, EN/DE parity check #544 [Docs] Docs CI: build-only PR check, link checker, EN/DE parity
• [Docs] Sweep stale version strings in docs (0.9.1 / 0.10.0 / 0.11.0) #546 [Docs] Sweep stale version strings

Quality of life / DX

• [Feature] system.spawn(ActorClass, name) overload — drop the Props ceremony for the common case #547 [Feature] system.spawn(ActorClass, name) overload
• [Feature] protected reply() on Actor #548 [Feature] protected reply() on Actor
• [Feature] Graceful-shutdown wiring: HTTP unbind + cluster leave via CoordinatedShutdown + runUntilTerminated() #549 [Feature] Graceful-shutdown wiring via CoordinatedShutdown + runUntilTerminated()
• [Docs] examples/README.md index + document the 47 orphaned examples #550 [Docs] examples/README.md index + document the 47 orphaned examples
• [Docs] Example pack: sharding entityRefFor, cluster deployment (compose/k8s), broker pipeline, DevTools federation #551 [Docs] Example pack: sharding, deployment, broker pipeline, DevTools federation
• [Feature] Decouple examples from the devtools harness #552 [Feature] Decouple examples from the devtools harness
• [Feature] DevTools UI: new panels — dead letters, EventStream viewer, resolved-config inspector, send-message action #553 [Feature] DevTools UI: new panels (post-[Feature] DevTools UI: migrate to Angular 22 + Apache ECharts (epic) #482, deliberately low priority)

Slimming

• [Feature] Delete the shadowed src/pattern/Status.ts (dead public API) #554 [Feature] Delete shadowed src/pattern/Status.ts
• [Feature] HttpBackendBase template class — deduplicate Fastify/Express/Hono request normalization #555 [Feature] HttpBackendBase template (enables a single-point fix for fix(http): request body-size limit parity across backends #357)
• [Feature] Explicit exports for cluster/http/io in the root barrel #556 [Feature] Explicit exports for cluster/http/io in the root barrel (with [Feature] Add subpath exports to package.json #414)
• [Feature] Decouple HttpManagement from src/crdt #557 [Feature] Decouple HttpManagement from src/crdt
• [Feature] Consolidate the four typed-spawn entry points #558 [Feature] Consolidate the four typed-spawn entry points
• [Feature] Stop committing example frontend build artifacts + single parameterized integration-test driver #559 [Feature] Stop committing example frontend bundles + parameterized integration driver

Positioning

• [Docs] "Why actor-ts" comparison page: vs Rivet, Restate, Cloudflare Durable Objects, Temporal, nact #560 [Docs] "Why actor-ts" comparison page (Rivet/Restate/DO/Temporal/nact)
• [Docs] Agent-memory framing: persistence + sharding as durable agent state #561 [Docs] Agent-memory framing (persistence + sharding as durable agent state — what Akka now markets as "Akka Memory")

Research notes for existing backlog priorities

• Pekko 1.2.x concentrates its feature investment in streams (majority of additions are new operators) — supports prioritizing [Feature] Akka-Streams DSL subset (SourceQueue, MergeHub, BroadcastHub) #147 (streams subset) when the L/XL threads come up.
• The claim "devtools UIs drive framework adoption" was adversarially refuted in our research round — finish [Feature] DevTools UI: migrate to Angular 22 + Apache ECharts (epic) #482 as planned, but further DevTools expansion ([Feature] DevTools UI: new panels — dead letters, EventStream viewer, resolved-config inspector, send-message action #553) is deliberately low priority.
• Verified demand signal: the dedicated Node/TS actor niche is underserved (nact stalled pre-clustering, last push 2023; Dapr took ~21 months to ship JS actors) — typed actors + clustering is the market gap this project already fills; [Docs] "Why actor-ts" comparison page: vs Rivet, Restate, Cloudflare Durable Objects, Temporal, nact #560 turns that into positioning.

Suggested immediate slice for the next minor: #527, #528 -> #529, #545, #539, #549, #547, #532 — with #530 started in parallel as a design thread.