Release v0.0.76

Overview

This release adds AI to the task palette and sharpens the caching and concurrency primitives. A new Dapr Conversation task (TaskType = 20) invokes an LLM/AI provider (OpenAI, Anthropic, Bedrock, …) provider-agnostically through the Dapr Conversation building block (#844). Function result caching gains vary-by headers — varyByHeaders / varyByHeaderPrefixes feed a varyKey(context) key-expression helper so one function keeps separate cache variants per header value (#839). Flow-level cache tuning is consolidated under a single attributes.config object (config.functionCache.ttlSeconds), the author-controlled TTL for the built-in instance functions (#846). Distributed resource locks become production-ready with idempotent release and automatic terminal cleanup (#840), plus a per-subInstance lock key for subflow terminal-outcome propagation (#845). Wizard states are now treated like normal states in the State/View functions (#838). The release also fixes author Content-Type preservation (#847), case-insensitive request-header normalization (#841), and init mappings support for publish (#848). This release runs on component schema 0.0.51.

---

Features

Dapr Conversation task — provider-agnostic LLM calls (#844)

A new task type, DaprConversation (TaskType = 20), invokes an LLM/AI provider through Dapr's Conversation building block — the same sidecar the other Dapr task types use. The provider and its credentials live in a domain-owned Dapr conversation component (e.g. openai); the task only references the component name and the messages.

{
  "attributes": {
    "type": "20",
    "config": {
      "componentName": "openai",
      "inputs": [
        { "role": "system", "content": "You summarize customer complaints." },
        { "role": "user", "content": "Summarize the complaint in two sentences." }
      ],
      "parameters": { "model": "gpt-4o-mini", "maxTokens": "512" },
      "temperature": 0.2,
      "scrubPII": true
    }
  }
}

• Required config is just componentName; inputs is an array of role/content messages, and parameters carries provider-specific string values (model, maxTokens, …).
• Optional metadata, contextId (stateful conversations), temperature, scrubPII, and timeoutSeconds (default 30).
• The runtime component is installed via the helm chart (vnext-helm-charts #28), so shipping a new provider needs no runtime redeploy.

Reference: issue #844, helm PR vnext-helm-charts #28 — see also the new Dapr Conversation Task component page.

Function cache vary-by headers (#839)

The function attributes.cache block gains two fields — varyByHeaders (exact request-header names) and varyByHeaderPrefixes (header-name prefixes) — that fold the named headers into the cache key. They form the header-name set for the varyKey(context) key-expression helper (the two lists are unioned), so a single read function keeps separate cache variants per header value without the runtime baking in any header convention. When the instance does not supply its own Instance.Data["varyBy"], these domain-declared fields are used.

Reference: issue #839 — see also Custom Functions → Cache.

Flow-level cache config object (#846)

Author-controlled, flow-scoped cache tuning is now consolidated under a single attributes.config object. Its first member, functionCache.ttlSeconds, is the TTL for this workflow's built-in instance functions (data, view, schema, …).

{ "config": { "functionCache": { "ttlSeconds": 120 } } }

• Repeat calls for the same instance are served from cache for the TTL window; when the instance changes, the cache drops and the next call is re-cached.
• The host default is 60s when config (or ttlSeconds) is absent or non-positive.
• The State Function is excluded — the platform manages its cache separately (host-side StateFunctionCache).

Reference: issue #846 — see also Workflow component → Config.

Distributed resource locks — idempotent release + automatic cleanup (#840, #845)

Start, state-level, and shared transitions can declare an optional resourceLock block — a distributed lock (Dapr lock.redis, via the Aether SDK) that prevents concurrent instances from mutating a shared resource (a seat, daily limit, account, …). It runs in the Manual profile only, is owned by the instanceId, and always carries a TTL.

{
  "key": "reserve-daily-limit",
  "target": "limit-reserved",
  "triggerType": 0,
  "versionStrategy": "Patch",
  "resourceLock": {
    "keyExpression": { "location": "./src/DailyLimitKey.csx", "code": "<base64>", "type": "L", "encoding": "NAT" },
    "action": "Acquire",
    "ttlSeconds": 300,
    "onConflict": "Abort"
  }
}

• The recommended model is Acquire on the check/entry transition and let the runtime release it: when the instance reaches a terminal state (Completed / Faulted / Cancelled), its locks are released automatically — no manual Release on every terminal transition.
• Release is idempotent/best-effort — LockDoesNotExist counts as success, a foreign lock is logged (not faulted), and lock cleanup never rolls back a successful business transition.
• A conflict on Acquire aborts the transition and returns HTTP 409 (the instance is marked faulted F); the caller retries. Extend is unreliable (no native Dapr extend) — size the TTL to cover the whole operation.
• Subflows now derive a per-subInstance lock key so terminal-outcome propagation stays correct across nested instances (#845).

Reference: issues #840, #845 — see also the new Resource Lock guide and Workflow component → Transition.

Wizard states behave like normal states in read functions (#838)

The State and View functions now treat Wizard states like normal states: authorization/role evaluation and the available-transition list are resolved the same way, so a wizard state's state/view response is consistent with any other state. This removes the special-casing that previously diverged wizard-state reads.

Reference: issue #838 — see also Workflow component → Wizard State ve View Davranışı.

---

Fixes

• Author Content-Type preserved without charset suffix — function output scripts that set content-type keep the author value verbatim (the ; charset=... suffix is no longer appended) (#847).
• Case-insensitive request-header normalization — transition request headers are matched case-insensitively, so header lookups no longer miss on casing differences (#841).
• Init mappings support for publish — the init/publish path handles mapping components correctly (#848).

---

Configuration Updates

Configuration for v0.0.76:

{
  "runtimeVersion": "0.0.76",
  "schemaVersion": "0.0.51"
}

Note: Schema version advances to 0.0.51 (from 0.0.50). The bump adds the DaprConversation task type "20" (#844), the function-cache varyByHeaders / varyByHeaderPrefixes fields (#839), the workflow-level attributes.config object (config.functionCache) (#846), and the transition resourceLock definition (#840). Update @burgan-tech/vnext-schema in your domain project before validating against this runtime.

Container images: published at tag 0.0.76 under ghcr.io/burgan-tech/vnext/*, Cosign-signed (keyless OIDC) with SBOM + provenance. Immutable digests are listed in the GitHub release.

---

Issues Referenced

• vnext #844 — Dapr Conversation (AI/LLM) task type.
• vnext #839 — varyKey cache-key helper + config-driven vary-by headers.
• vnext #846 — Consolidate flow-level functionCache under a config object.
• vnext #840 — Resource lock: idempotent release + automatic terminal cleanup.
• vnext #845 — Retryable 503 + per-subInstance lock key for terminal-outcome propagation.
• vnext #838 — Treat wizard states like normal states in State/View functions.
• vnext #847 — Preserve author Content-Type without charset suffix.
• vnext #841 — Normalize request headers case-insensitively.
• vnext #848 — Init mappings support for publish.
• vnext-helm-charts #28 — Dapr Conversation helm component.

---

Summary

• Dapr Conversation task (Type 20) invokes an LLM/AI provider provider-agnostically through the Dapr Conversation building block; the provider is a domain-owned Dapr component.
• Function cache vary-by headers: varyByHeaders / varyByHeaderPrefixes feed the varyKey(context) helper for per-header cache variants.
• Flow-level config.functionCache.ttlSeconds: author-controlled TTL for built-in instance functions (host default 60s; State Function excluded).
• Distributed resource locks are production-ready: idempotent release, automatic terminal cleanup, HTTP 409 on conflict, and a per-subInstance lock key for subflows.
• Wizard states are read like normal states in the State/View functions.
• Fixes: author Content-Type preserved, case-insensitive header matching, init mappings support for publish.
• Schema is 0.0.51.

---

vNext Runtime Platform Team Released July 27, 2026