Cloudflare CI SDK, Rspack 2.0 ESM, Ollama fixes — Dev Signal
Cloudflare CI SDK, Rspack 2.0 ESM, Ollama fixes
Share:
Tool of the Week
Cloudflare ships CI SDK built on Workflows
Define CI/CD pipelines in TypeScript instead of YAML, trigger directly on artifact push, cache dependencies across steps with native observability in Workflows dashboard.
Eliminates event subscription/queue boilerplate for triggering CI jobs and replaces YAML with TypeScript for full customization. Dependency caching and parallel step execution reduce pipeline latency while platform teams can now share one CI definition across all customer repos.
Replaces GitHub Actions for Cloudflare-hosted code. Requires: Artifacts repo, wrangler config with events field, @cloudflare/ci SDK import, TypeScript workflow definition. Ready to try now for new Cloudflare-native projects; migration from Actions is feasible but not a drop-in replacement.
“We have stitched the store, build, and deploy steps together with the CI SDK, built on Cloudflare Workflows”
“A CI/CD pipeline is just a Workflow”
“Instead of YAML, you can define your CI/CD pipeline in Typescript for greater customization and configurability”
“you can now kick off a CI job directly on push instead of configuring an event subscription, a queue, and a queue consumer”
“Dependency caching reduces the latency of your CI/CD pipeline since every CI step won't need to rerun the install”
“Each CI run surfaces as a Workflow instance so you can view its step-by-step execution and observability directly in the Workflows dashboard”
Get issues like this in your inbox — free, every weekday.
Quick Signals
Vercel Blob adds consistent reads for private storage
Pass `useCache: false` to bypass CDN caching on blob reads, guaranteeing read-after-write consistency within 60 seconds of overwrites.
Agent memory files, session transcripts, and real-time reports now have guaranteed freshness without manual cache-busting. Developers building stateful AI systems can rely on immediate consistency instead of eventual coherence.
Replaces manual cache-busting patterns. Requires `@vercel/blob@2.6.1+` and willingness to accept slower reads and Fast Origin Transfer costs for critical paths. Ready now—add flags surgically to hot paths only.
“Pass `useCache: false` to `get()` or `presignUrl()` for a read that reflects the latest write”
“readers might see the cached version for up to 60 seconds”
“These reads bypass the CDN, take longer than cached reads, and incur Fast Origin Transfer”
v0.32.1-rc0 patches MLX model cache memory leak, stabilizes Gemma 4 tool-response continuations, and adds working directory context to agents.
Memory leak fixes directly reduce production resource costs for long-running agent deployments. Improved tool calling reliability cuts failures in multi-turn reasoning workflows.
Drop-in patch for v0.32.0 users. Cache leak fix requires upgrade if you run persistent agents. Tool calling improvements benefit Gemma 4 users immediately; others see no breaking changes. Worth upgrading now if you've seen memory creep in multi-request sessions.
Data Point
Meta-router composes operations under cost budgets
Learned policy router selects execution paths (decompose, retrieve, code, delegate, verify) from task text using regularized logistic heads, achieving 100% success on test tasks at 43% lower cost than static routing.
Agentic systems currently waste budget on fixed workflows; routing decisions per task let you optimize for latency, cost, or success rate without rebuilding pipelines. The benchmark provides reproducible evaluation for your own routing strategies.
Replaces hard-coded routing logic with learned policies, but requires: labeled task→operations data (216+ examples), outcome verification infrastructure, and cost accounting per operation. Proof-of-concept only—lexical generalization fails on unseen tasks (75.9% vs 93.5% static on challenge split). Not production-ready; useful for research prototyping.
“A controller may answer directly, decompose a request, retrieve evidence, execute code, delegate to a specialist, or verify an intermediate result”
“achieved 100% success versus 93.5% for strong static and fixed workflows, with 43% lower cost than the static policy”
“The gap identifies lexical generalization, rather than route execution, as the principal limitation”
“These results establish a reproducible testbed and a bounded proof of concept, not evidence of live-LLM performance”
MLX engine now auto-enables speculative decoding via MTP head; streaming format aligns with OpenAI's wire protocol for easier client migration.
Speculative decoding on Apple Silicon reduces inference latency for local deployments. Matching OpenAI's streaming format eliminates format translation logic when switching providers or running local alternatives.
Replaces manual speculative decoding setup for Qwen3.5 on MLX; requires upgrading to v0.32.6-rc0. Streaming change is transparent if you parse chunks generically, but verify finish_reason and usage handling in your streaming consumer. Image generation users must stay on 0.32.5 until feature returns. Worth upgrading for Apple GPU users and OpenAI client compatibility; treat RC as release-candidate-ready.
“Qwen3.5 is faster on Apple GPUs: the MLX engine now uses the model's MTP head for speculative decoding automatically”
“/v1/chat/completions streaming now matches OpenAI's wire format: role only on the first chunk, finish_reason on its own chunk, and usage in a separate chunk with stream_options.include_usage”
“Experimental image generation has been temporarily removed. Continue using 0.32.5 for image generation support”
Android Studio Quail 2 enables parallel AI agent conversations
Agent Mode redesign removes sequential task bottleneck, allows simultaneous conversations across multiple LLMs via Android Bench benchmarking; LeakCanary integration speeds heap analysis 5x by offloading to dev machine.
Eliminates idle time waiting for agent task completion, letting you parallelize refactoring, debugging, and documentation work. LeakCanary integration cuts memory leak detection time from constrained test devices to your development machine, reducing jank during debugging.
Replaces sequential agent workflows and manual heap analysis. Requires Android Studio Quail 2 stable (now available); LeakCanary integration automatic. Worth adopting immediately if you debug memory leaks or use Agent Mode regularly—concrete 5x perf gain on leak tracing is measurable.
“Waiting for the Android Studio agent to finish a task before you can ask another question or initiate a separate task in Agent Mode is a bottleneck of the past.”
“leak tracing is up to five times faster and jank-free”
“Agent Mode has been completely redesigned in Quail 2, delivering improved performance, more effective task decomposition, and enhanced tool support”
Pure ESM bundler with 10% faster builds and @rspack/dev-server shrunk from 192 deps to 1 — swap webpack config compatibility for cleaner supply chain.
Rspack 2.0 eliminates the dependency bloat plaguing Node toolchains while maintaining webpack API parity, reducing install friction and build times on large projects. The ESM-first shift and persistent cache optimization directly speed iteration cycles on React/TypeScript codebases.
Replaces webpack for new projects; migrate from 1.x if you're already using Rspack. Requires Node 20.19+ or 22.12+ (drops v18). Worth trying now if your team values supply-chain hygiene and bundle performance—the 95% webpack config compatibility removes config rewrite risk, but ecosystem maturity lags webpack and Vite.
“overall build performance is around 10% faster than Rspack 1.7 and as much as 100% faster than 1.0”
“production builds with persistent cache dropped from 5.6 seconds on 1.0 to 1.4 seconds on 2.0”
“The @rspack/dev-server package went from 192 dependencies down to 1, and its install size shrank from 15 MB to 1.4 MB”
“@rspack/core is now published as a pure ESM package”