* feat: add scheduled job deletion with configurable retention period
Extends delete_after_use with delete_after_secs to enable configurable
retention periods for job args/result/logs. At completion, jobs can be
scheduled for future deletion via a new job_delete_schedule table,
processed by a monitor task. Supports per-script, per-flow, and
per-flow-step configuration. Backward compatible.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat: add integration tests, revert query! macros, fix review issues
- Add integration tests for resolve_delete_after_secs, schedule_job_deletion,
flow-level and module-level delete_after_secs, backward compat
- Revert sqlx::query() back to sqlx::query!() macros for compile-time safety
- Regenerate sqlx offline cache
- Fix FlowModule/NewScript/FlowValue constructions in all test files
- Fix autoscaling_ee.rs for updated script_path_to_payload return type
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: update ee-repo-ref.txt for autoscaling_ee fix
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: gate cleanup_scheduled_job_deletions behind enterprise feature
Prevents dead_code warning (which CI treats as error via -D warnings)
when compiling without enterprise feature.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: regenerate sqlx cache after merge with main
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: address review feedback on scheduled deletion
- Monitor: roll back transaction on any cleanup error so schedule rows
survive for retry on next cycle (instead of best-effort then discard)
- Migration: add FK with ON DELETE CASCADE to job_delete_schedule.job_id
to prevent orphan rows when jobs are deleted through other means
- Simplify bool-to-Option conversion with .then_some(true)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor: stop setting delete_after_use alongside delete_after_secs
No mixed-version deployment scenario exists, so delete_after_secs alone
is sufficient. The backend's resolve_delete_after_secs handles
(None, Some(secs)) correctly without needing delete_after_use set.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor: remove delete_after_use from public API surface
Remove delete_after_use from OpenAPI spec, API client, runtime client,
and workspace export. Only delete_after_secs is exposed going forward.
The field remains in Rust backend types with #[serde(skip_serializing)]
for backward-compatible deserialization of existing scripts/flows that
were saved with delete_after_use: true.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: update ee-repo-ref to 1d4b7a31fc115d6aba8640f7cd3fd5a01abe6806
This commit updates the EE repository reference after PR #519 was merged in windmill-ee-private.
Previous ee-repo-ref: 9eba09a13b778caafc6ae65098b90e53c91984d3
New ee-repo-ref: 1d4b7a31fc115d6aba8640f7cd3fd5a01abe6806
Automated by sync-ee-ref workflow.
* fix: regenerate system prompts, remove unused import
- Regenerate auto-generated system prompts after openflow schema change
- Remove unused serde_json::json import in test file (CI -D warnings)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: insert dummy v2_job row in schedule tests for FK constraint
The job_delete_schedule table has a FK to v2_job, so tests need a
real v2_job row before inserting into the schedule table.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: trigger CI re-run
* fix: remove heavy flow integration tests to avoid CI worker contention
The flow integration tests spawn workers that compete for CPU with
the existing relock_skip tests under --test-threads=10, causing
consistent 60s timeouts in CI. Keep only the lightweight unit tests
and DB integration tests.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: restore correct ee-repo-ref for our branch
The ref was overwritten to main's EE ref during a rebase. Restore to
our branch's EE commit that includes the autoscaling tuple fix.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: retrigger CI on fresh runner
* fix: remove FK constraint from job_delete_schedule to unblock CI
The FK with ON DELETE CASCADE to v2_job may have caused performance
overhead during test DB setup (each sqlx::test creates a fresh DB
with all migrations). Remove the FK — orphan schedule rows are
harmlessly cleaned by the monitor.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* ee-ref
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* feat: add PDF input support to AI agent with user_attachments field
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* test: add integration tests for PDF input and backward compat
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat: add ContentPart::File variant for PDF support across all providers
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor: address review feedback on PDF support
- Extract parse_data_url_bytes and mime_to_document_format helpers in Bedrock
- Add is_document_mime helper in ai_types for centralized MIME routing
- Extract s3_object_to_content_part helper to deduplicate image_handler/openai
- Rename AnthropicImageSource to AnthropicBase64Source
- Derive Bedrock DocumentFormat from MIME type instead of hardcoding Pdf
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: merge user message and attachments into single message for Bedrock
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: prevent AI agent tool jobs from becoming zombies on cancellation
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* kill tool tasks on cancel timeout
* fix: address review feedback and update sqlx cache
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor: rename AnthropicPlatform to AIPlatform for generic vertex support
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add Vertex AI support for Google Gemini models
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update platform doc comments to be provider-generic
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update googleai default models to latest gemini 2.5 and 3
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: move AIPlatform to windmill_common to avoid duplication
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: hugocasa <hugo@casademont.ch>
* feat: workflow-as-code v2 with @task decorator API
Replace ctx.step("name", "script") API with @task decorators where
functions are called directly. Users no longer need to pass WorkflowCtx
or use string-based step names/script paths.
Python: @task decorator with contextvars-based implicit context
TypeScript: task() wrapper with module-level context variable
Parsers: detect @task function calls instead of ctx.step() calls
Worker: updated wrappers to set implicit context
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: WAC v2 checkpoint/replay with _executing_key child dispatch
- Rust-side orchestration: parent dispatches child jobs, suspends, resumes on completion
- _executing_key in checkpoint tells child which step to execute directly
- task() throws StepSuspend(mode="step_complete") after executing target step
- result_processor handles child completion and updates parent checkpoint
- WacGraph.svelte for runtime execution visualization
- Sequential and parallel workflows tested end-to-end
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: WAC v2 bundle cache, globalThis ctx sharing, description optional
- Disable bun bundle caching for WAC v2 scripts (wrapper needs
windmill-client from node_modules, not available in bundle mode)
- Use Reflect.set/get(globalThis, "__wmill_wf_ctx") to share workflow
context across dual module instances (wrapper vs user script)
- Never-resolving thenable for non-matching steps in child job mode
prevents Promise.all race conditions
- Make description field optional in NewScript API (defaults to "")
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add step() primitive for inline checkpointed steps
step() executes a function inline (no child job) and persists the result
to the checkpoint. On replay, the cached value is returned — ensuring
deterministic behavior for non-deterministic operations like Date.now()
or Math.random().
- TypeScript: step(name, fn) — executes inline, throws StepSuspend with
mode "inline_checkpoint" to persist before continuing
- Rust: InlineCheckpoint variant in WacOutput, saves to checkpoint and
resets running=false for immediate re-pickup (no zombie wait)
- Shared step counter between task() and step() via _allocKey()
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add Python WAC v2 support with task(), step(), workflow()
- Python SDK: WorkflowCtx with _executing_key child mode, _alloc_key
shared counter, _run_inline_step for step(), _execute_directly and
_never_resolve for child mode, step() async function
- Python executor: WAC v2 detection, checkpoint.json writing, WAC
wrapper.py generation calling _run_workflow(), post-execution hook
into shared handle_wac_v2_output()
- Make handle_wac_v2_output pub so both bun and python executors share
the same dispatch/suspend/inline-checkpoint logic
- 17 Python tests covering dispatch, replay, parallel, conditional,
inline checkpoint, and child mode
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update sqlx prepared queries
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: WacGraph Tooltip→Popover, simplify wacToFlow parsers
- Fix type error: Tooltip doesn't accept text snippet, use Popover
- Extract shared helpers for task matching and block collection
- Replace linear tasks.find() with Map lookups
- Remove mutable module-level counter
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: Box::pin WAC v2 output handler to prevent stack overflow
handle_python_job's async state machine was too large when combined
with handle_wac_v2_output. Box::pin heap-allocates the future.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: merge WAC v1 and v2 task decorators to preserve backward compat
The v2 @task decorator was shadowing the v1 one, breaking WAC v1
scripts that rely on HTTP-based dispatch via /workflow_as_code/ API.
The merged decorator handles three modes:
- v2: inside @workflow context → checkpoint/replay dispatch
- v1: WM_JOB_ID set, no @workflow → HTTP API dispatch + wait_job
- standalone: no Windmill env → execute function body directly
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: skip no_main_func detection for WAC v2 scripts in TS and Python parsers
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: prevent empty/noop dispatch causing infinite requeue loop
- Validate steps.len() > 0 in WAC dispatch handler (issue 3)
- Replace noop StepSuspend throw with never-resolving promise so it
can't reach the backend as an empty dispatch (issue 4)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: Python task wrapper now converts positional args to kwargs in v2 mode
Previously only **kwargs were passed to _next_step(), silently dropping
positional arguments. Extract shared _merge_args() helper used by both
v1 and v2 paths.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: replace unwrap() with proper error propagation in WAC arg serialization
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: add workspace_id filter to v2_job queries in WAC dispatch
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: prevent race condition in WAC child dispatch
Restructure dispatch to save checkpoint + suspend parent + seed child
checkpoints in a single transaction BEFORE pushing child jobs. This
ensures a fast child can't complete before the parent is suspended.
Also wrap InlineCheckpoint save + running reset in a transaction to
prevent corrupted state on crash.
Use ULID for pre-generated child job IDs (consistent with rest of API).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: include step key and child job ID in WAC error propagation
Move step_key lookup before the success check so failed child errors
include which task failed, the child job ID, and the original error.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: document WAC determinism contract and step dispatch semantics
- Document that workflow functions must be deterministic across replays
- Document that WacStepDispatch.script/args are metadata, not dispatch targets
- Add comments on counter-based key allocation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: tighten WAC v2 detection to reduce false positives
Replace naive substring matching with line-aware checks that skip
comments and look for specific patterns:
- TS: import from "windmill-client" containing workflow/task
- Python: @workflow and @task decorators with wmill import
Extracted shared helpers in wac_executor.rs used by both executors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: show failed steps in WacGraph when workflow completes with errors
When flowDone is true and a pending step isn't in completedSteps,
mark it as 'failed' instead of 'running'. The failed state CSS and
XCircle icon were already defined but never triggered.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: unsuspend and fail parent when WAC child push fails
Previously if a child push failed mid-batch, the parent remained
suspended with suspend = num_steps but fewer children, hanging until
the 14-day timeout. Now the push loop catches errors and unsuspends
the parent before returning the error.
Also adds source hash validation: if the script content changes between
replays, the job fails with a clear error instead of silently feeding
stale checkpoint data into wrong steps.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: clear suspend_until when unsuspending WAC parent
Set suspend_until = NULL alongside suspend = 0 in both the child
failure and all-children-complete paths, so the parent doesn't rely
on subtle pull query invariants to be re-picked-up.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* test: add exhaustive edge case tests for WAC v2 SDK
fix: make TS task wrapper non-async to fix unawaited task flush
The async wrapper caused microtask-based thenable auto-resolution that
fired .then() and threw StepSuspend before _flushPending() could capture
unawaited steps — making the flush mechanism completely broken. Now the
thenable is returned directly without async wrapping. Backward compatible
with v1 (all code paths still return awaitables).
Tests added (59 TS + 66 Python) covering: full sequential lifecycle,
step after parallel, parallel after parallel, conditional on step result,
empty/single-task workflows, 10+ steps, falsy value preservation, inline
steps, mixed step/task, unawaited flush, child mode with parallel,
key determinism, large parallel groups, and complex mixed patterns.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: atomic checkpoint updates to prevent parallel child race condition
Replace read-modify-write pattern in handle_wac_child_completion with
atomic SQL operations:
- completed_steps merged via jsonb_set(... || jsonb_build_object(...))
so concurrent children on different workers don't overwrite each other
- suspend counter decremented atomically with RETURNING to determine
"all done" condition (instead of checking completed_steps in memory)
- suspend_until cleared in the same atomic decrement statement
Before this fix, two parallel children completing simultaneously could
both load the same checkpoint, each add their step, and save — the
second write would overwrite the first, silently losing a child result
and leaving the parent suspended forever.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: cancel already-pushed children on partial WAC dispatch failure
When pushing child jobs sequentially, if pushing child N fails, children
1..N-1 are already running. Previously the error handler only unsuspended
the parent, leaving orphaned children that would complete and corrupt the
checkpoint state (decrementing suspend on an already-unsuspended parent,
potentially causing duplicate step execution on re-run).
Now on partial failure:
1. Cancel all already-pushed children (prevents them from completing
and corrupting checkpoint state)
2. Clear pending_steps from checkpoint (so parent doesn't think
children are outstanding on re-run)
3. Then unsuspend parent (so the error propagates)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: skip WAC duration write and child check for non-WAC parents
The duration write to workflow_as_code_status was running for every
non-flow child with a parent (error handlers, success handlers,
run_script children), even though it was only intended for WAC jobs.
Add WHERE workflow_as_code_status IS NOT NULL to skip non-WAC parents
entirely. Piggyback RETURNING pending_steps.job_ids on the same query
so WAC v2 child completion needs zero extra DB round-trips on the
success path.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: seed child checkpoint in same transaction as push
The child checkpoint insert was happening before the child job was
pushed, violating the FK constraint on v2_job_status. Move it into
the push transaction so the job row exists and the child can't be
picked up before its checkpoint is ready.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: set running=false when WAC parent suspends for child dispatch
The parent job kept running=true after suspending, so workers wouldn't
pick it up when children completed and suspend reached 0. The parent
only advanced when the zombie job detector reset it (~90s). Now the
dispatch suspend sets running=false so the parent is immediately
eligible for pickup.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: WAC parent suspend/unsuspend lifecycle
Keep running=true when suspending the parent so the normal pull query
(WHERE running=false) never picks it up. Keep suspend_until non-null
when decrementing suspend to 0 so the suspended pull query
(WHERE suspend_until IS NOT NULL AND suspend<=0) picks it up.
Previously: setting running=false caused infinite restart loops because
the normal pull query has no suspend check and would immediately re-pick
the parent. Clearing suspend_until on the last child prevented the
suspended pull from ever seeing it, requiring the 90s zombie detector.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add approval primitive, flow child completion, timeline fixes for WAC v2
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add error propagation, task options, sleep, and parallel for WAC v2
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* test: fix python SDK tests to use name-based keys and add new test coverage
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: address WAC v2 review findings (sleep timing, error marker, atomicity)
- Fix sleep using suspend=1 instead of 0 to enforce actual delay
- Add approval/sleep resume injection to Python executor
- Fix TS SDK concurrency_limit mapping (was reading wrong property)
- Namespace error marker as __wmill_error to avoid user data collision
- Wrap child completion SQL in transaction for atomicity
- Decrement suspend even when step key is missing (prevents hang)
- Expand TASK_RE to handle export const, let, var, generics
- Validate step key uniqueness before dispatch
- Log warning on checkpoint deserialization failure
- Remove unimplemented delete_after_use from SDKs
- Add TaskError exception class to Python SDK with diagnostic context
- Fix extra positional args handling and add functools.wraps
- Improve getParamNames to handle typed/destructured params
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* sqlx
* sqlx
* test: add WAC v1 e2e integration tests for TS and Python
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: revert fake test versions in typescript-client
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove unused WacGraph component and strip wacToFlow to isWorkflowAsCode
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: extract shared approval/sleep resume logic into wac_executor
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: extract google ai logic to windmill-common and use native gemini api in chat proxy
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: use x-goog-api-key header for google ai non-chat requests
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: transform gemini models response to openai format and use correct auth header
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: skip thought parts from gemini thinking models in sse stream
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Revert "fix: skip thought parts from gemini thinking models in sse stream"
This reverts commit dfa01d282c.
* fix: handle tool calls and sanitize schemas in gemini chat proxy
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* refactor: move Gemini→OpenAI response conversion to windmill-common
Extract streaming and non-streaming Gemini response conversion into
shared functions in ai_google so the API proxy and worker use the same
logic instead of duplicating format translation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: review fixes for google ai refactor
- Remove duplicate parse_data_url from worker utils, use shared version
from windmill_common::ai_google in both google_ai and anthropic providers
- Improve error diagnostics in google.rs by including HTTP status code
in error messages from Gemini API responses
- Change GeminiToolCallEvent::into_extra_content to instance method
to_extra_content using &self
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: deduplicate worker Gemini message conversion using pre-flight pattern
Replace the worker's `convert_messages_to_gemini` and
`convert_content_to_parts_with_s3` (~130 lines) with the existing
pre-flight pattern: `prepare_messages_for_api` converts S3 objects to
data URLs, then the shared `openai_messages_to_gemini` handles the rest.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: hugocasa <hugo@casademont.ch>
* worker: support AIAgent tools in AI executor
* worker: complete nested AIAgent tool execution path
* worker: inline AIAgent tool schema usage
* fix agent action
* frontend: add AI Agent as tool type in flow builder
Add the ability to insert a nested AI Agent as a tool within another
AI Agent step. Includes type definitions, factory function, graph icon,
insert/event wiring, and a dedicated editor component.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove AiAgentToolEditor, reuse FlowModuleComponent for AI agent tools
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: populate all input transforms for nested AI agent tools
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: avoid missing v2_job_status error for nested AI agent tools
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* sqlx
* nit
* refactor: cleanup nested AI agent tool implementation
- Add max nesting depth guard (5) on parent chain traversal
- Reject 3+ level nesting explicitly with clear error message
- Remove unnecessary flow_step_id tuple scaffolding in tool dispatch
- Consolidate get_value() calls using borrow in first match
- Replace unsafe `as unknown as FlowModule` casts with agentToolToFlowModule()
- Simplify toolKind ternary chain with .includes() lookup
- Fix leftover over-indentation from tuple removal
- Remove duplicate doc comment on is_completed_input_transform
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: pass flow_step_id and flow_job_id overrides to run_agent for nested AI agents
For nested AI agent tools, job.flow_step_id is None and job.parent_job
points to the parent agent instead of the flow. This caused memory
read/write and flow context resolution to silently fail.
handle_ai_agent_job already computes the correct flow_step_id (via
runnable_path fallback) and flow_job_id (via parent chain traversal).
This change threads those values through run_agent and
ToolExecutionContext so all downstream consumers use the correct IDs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* cleaning
* cleaning
* move const
* fix
* refactor: replace defaultToAi boolean with allowedAiTransforms whitelist
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* refactor: propagate root_job at push time, remove flow_job_id_override
Instead of threading flow_job_id_override through run_agent and
get_flow_context, propagate root_job and flow_innermost_root_job
when pushing tool jobs so nested AI agents can find the flow
job naturally via the existing job fields.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: simplify nested AI agent parent chain walk-up
Replace the generic depth-limited loop with a single-level check since
only flow → agent → nested agent tool is supported. Remove
MAX_AGENT_NESTING_DEPTH constant and flatten the module lookup.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: reject 3+ level nested AI agent tools before job creation
Check at the parent agent level whether a nested AIAgent tool contains
AIAgent sub-tools. If so, return a fatal error immediately, preventing
the sub-job from being created and avoiding retry loops.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: resolve deadlock in nested AI agent tool execution
Replace channel forwarding with inline DB writes for tool job
completions. Nested agents used bounded(1) channels where a sub-tool's
forwarded result would fill the parent channel, leaving no room for the
agent's own completion — causing a deadlock. Writing directly via
add_completed_job/add_completed_job_error bypasses the channel entirely.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: extract object store code into windmill-object-store crate with filesystem backend
Consolidate all object_store-dependent code from windmill-common into a new
windmill-object-store crate. Add a filesystem-backed object store implementation
using LocalFileSystem for dev/testing without cloud credentials. Includes 30
comprehensive tests covering render_endpoint, lfs_to_object_store_resource,
duckdb_connection_settings, error mapping, and filesystem-backed integration tests.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* all
* all
* all
* all
* fix: fix raw_app hardcoded path, add missing ObjectStoreResource import, and add tests
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: move S3ModeFormat to windmill-types, make windmill-parser-sql optional, restore debug logs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* all
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add AWS Bedrock session token support in API and worker
* Add Bedrock auth mode integration tests for AI agents
* Split Bedrock integration test env vars for IAM and session creds
* cleaning
* Add masked Bedrock bearer-token debug info logs
* Revert "Add masked Bedrock bearer-token debug info logs"
This reverts commit 6b2fc5e7c2d5b1c6db81f416a4439941a084108c.
* cleaning
* feat(ai): support 1M context window for Anthropic resources
Add `enable_1m_context` boolean field to Anthropic resource configuration.
When enabled (and not using Vertex AI), sends the `anthropic-beta: context-1m-2025-08-07`
header in both the API proxy layer and the AI agent worker layer.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ai): add serde alias for enable_1M_context DB field name
The resource_type schema uses `enable_1M_context` (uppercase M) but
serde only matched `enable_1m_context` and `enable1mContext`, causing
the field to always deserialize as false.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add prompt caching support for Anthropic API
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* exclude vertex
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* fix: parse Python datetime.datetime and datetime.date type annotations correctly
The Python parser only matched ExprKind::Name for type annotations, so
`datetime.datetime` (an Attribute expression) silently fell through to
Typ::Unknown and no datetime picker was shown in the UI.
- Extend parse_expr to resolve `datetime.*` attribute access (alongside
the existing `wmill.*` handling)
- Add Typ::Date variant for `datetime.date` → JSON schema format "date"
- Update python worker to import and convert `date.fromisoformat()`
- Update argSigToJsonSchemaType, AI types, schema validation, and SQL
datatype wasm for the new Date variant
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* all
* all
* all
* all
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: extract windmill-dep-map crate for parallel api/worker compilation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: resolve WebhookShared type mismatch and missing enterprise propagation
- Make windmill-api webhook_util re-export from windmill-common instead of
duplicating types, fixing Extension<WebhookShared> mismatch between
windmill-store and windmill-api
- Add windmill-api-jobs/enterprise to windmill-trigger enterprise feature
so check_license_key_valid is available when trigger subcrates enable
enterprise on windmill-trigger
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: stop trigger features from unconditionally enabling enterprise
Move enterprise propagation for all trigger subcrates from individual
trigger feature definitions to the enterprise feature itself, so
enterprise is only enabled when explicitly requested.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove unused pub use re-exports and disable CI cargo cache
- Remove unused re-exports from windmill-worker/src/lib.rs:
trigger_dependents_to_recompute_dependencies, handle_job_error,
and unused bun/otel items
- Fix callers to use direct module paths instead
- Add windmill-dep-map as dev-dependency for tests
- Disable cargo cache in backend-check CI (faster from-scratch builds)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: restore bun re-exports used by tests
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* all
* chore: re-enable cargo cache for check_ee_full CI job
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: remove $schema field from Google AI output schema requests
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* test: add $schema field to all output schema integration tests
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: remove $schema field from Google AI tool parameter schemas
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* test: add workspace script tool test for AI agents
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* feat: add token usage tracking to AI agent output
Add TokenUsage struct to track input/output/cache tokens from AI providers.
Currently implemented for Bedrock provider, with infrastructure in place
for other providers. Usage is included in the AI agent result alongside
output and messages when available.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: add token usage extraction for Anthropic provider
Extract usage from message_delta SSE event and convert to TokenUsage.
Includes input_tokens, output_tokens, cache_read_input_tokens, and
cache_creation_input_tokens (mapped to cache_write_input_tokens).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: add token usage extraction for Google AI/Gemini provider
Extract usage from usageMetadata in Gemini SSE events and convert to TokenUsage.
Maps promptTokenCount -> input_tokens, candidatesTokenCount -> output_tokens,
totalTokenCount -> total_tokens.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: add token usage extraction for OpenAI Responses API provider
Extract usage from response.completed SSE event and convert to TokenUsage.
Maps input_tokens, output_tokens, and total_tokens directly.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: add token usage extraction for Azure OpenAI / Chat Completions API
Add stream_options.include_usage to request and parse usage from final
SSE chunk for providers using the standard OpenAI Chat Completions API
(Azure OpenAI, Mistral, DeepSeek, Groq, TogetherAI, CustomAI).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: cleanup token usage tracking - remove unused Image usage field and accumulate across iterations
- Remove unused `usage` field from ParsedResponse::Image variant
- Add TokenUsage::accumulate() method to sum usage across agent iterations
- Accumulate input/output/total/cache tokens instead of replacing with last iteration
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* fix: remove verbose debug logging from AI providers
Remove tracing::info!("[debug] ...") statements that were too verbose
for production. These logged raw events on every streaming event.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* feat: add retry mechanism for OpenAI-compatible providers without stream_options support
Some OpenAI-compatible providers don't support the stream_options parameter
for usage tracking. This adds a retry mechanism that:
- First attempts the request with stream_options.include_usage
- If it fails with 400 and error mentions stream_options/include_usage,
automatically retries without the parameter
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* fix: remove unnecessary text parsing overhead in image response handlers
Revert debugging changes that read response as text before parsing JSON.
Using response.json() directly is more efficient.
Co-Authored-By: Claude <noreply@anthropic.com>
* refactor: centralize TokenUsage conversion with constructor methods
Add new(), from_input_output(), and with_cache() constructors to TokenUsage
to eliminate duplicate conversion logic across providers. Also fixes potential
truncation in Bedrock cache token conversion by using i32::try_from with
fallback to i32::MAX.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* refactor: simplify Anthropic usage extraction and add Default derive
- Use idiomatic `if let` pattern instead of `is_some()` check for usage extraction
- Add Default derive to OpenAIChatUsage for consistency with other usage structs
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: use saturating_add to prevent overflow in token accumulation
In long-running agents with many iterations, token counts could
potentially overflow. Using saturating_add ensures values cap at
i32::MAX instead of wrapping around.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* better claude
* nit
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* fix: handle empty strings in AI resource fields via serde deserializer
Add `empty_string_as_none` deserializer that converts empty strings to None
during deserialization. Applied to base_url, api_key, region, and AWS
credential fields in AIStandardResource and ProviderResource.
This fixes the "relative URL without a base" error when creating Anthropic
resources with empty base_url fields.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* nit
* nit
* nit
* cleaning
* cleaning
* cleaning
* cleaning
* fix: apply empty_string_as_none deserializer to api_key field
Consistent with other fields in ProviderResource, empty strings are now
deserialized as None for the api_key field.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* check endpoint
* use flag for cred check on worker
* use sdk for auth chat side
* cleaning
* cleaning
* also handle GET
* Add session token support in BedrockClient::from_credentials
Users with temporary STS credentials need session tokens for AWS
authentication. This adds the optional session_token parameter to
support these use cases.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Cache AWS SDK config loading to avoid repeated I/O
aws_config::load_defaults() performs environment variable lookups and
potentially file I/O on each call. Cache the result in a static OnceCell
to improve performance on high-frequency Bedrock requests.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add debouncing to Bedrock credential check button
Prevent rapid clicks from spawning multiple concurrent flow preview
jobs for the worker credential check.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Remove orphaned SigV4 signing comment
The manual SigV4 signing code was removed in favor of the AWS SDK,
but this documentation comment was left behind.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Remove verbose tracing info and change-related comments
- Remove per-request auth method info logs (too verbose for production)
- Simplify from_env log to single debug-level message
- Remove comments describing future changes rather than current behavior
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* cleaning
* Fix Bedrock feature flag compilation warnings
Gate AWS-specific imports, struct fields, and methods behind the bedrock
feature flag to eliminate dead code warnings when building without the
bedrock feature enabled.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* bedrock flag
* cleaning
* Move BedrockCredentialsCheck from edit drawer to add drawer
The component was incorrectly shown when editing resources instead of
when adding new ones. Moved it from ResourceEditor.svelte to
ApiConnectForm.svelte.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* cleaning
* cleaning
* make aws-config optional
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Co-authored-by: Ruben Fiszel <ruben@windmill.dev>
* refactor(ai): use responses API for OpenAI/Azure, create 'other' provider for completion endpoint
- Created new 'other.rs' provider using the OpenAI-compatible completion endpoint
- Refactored 'openai.rs' to use the Responses API for both text and image output
- Updated query_builder to route OpenAI/AzureOpenAI to OpenAIQueryBuilder
- All other providers (Mistral, DeepSeek, Groq, etc.) now use OtherQueryBuilder
- Updated OpenRouter to delegate to OtherQueryBuilder instead of OpenAIQueryBuilder
This prepares the codebase for adding websearch tool support using the Responses API.
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* feat(ai): add websearch as AI agent tool type
- Added WebsearchToolValue to ToolValue enum in flows.rs
- Updated all pattern matches to handle websearch tool type
- Added has_websearch parameter to run_agent function
- Websearch tools don't require additional configuration
This prepares the backend for provider-specific websearch implementations.
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* feat(frontend): add websearch as tool option in AI agent UI
- Added WebsearchTool type and utility functions to agentToolUtils.ts
- Added "Web Search" option in tool insertion menu (InsertModuleInner)
- Updated NewAIToolNode to handle pickWebsearchTool event
- Updated AIToolNode to render websearch tools properly
- Updated FlowModuleSchemaMap to create websearch tools
Users can now add websearch tools to AI agents through the UI.
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* feat(ai): implement websearch for OpenAI provider
- Added has_websearch parameter to BuildRequestArgs
- OpenAI provider now adds web_search tool to requests when enabled
- Uses OpenAI Responses API web_search tool type
- Websearch tool is added before other custom tools in the request
Implements websearch functionality for OpenAI and Azure OpenAI providers.
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* feat(ai): implement websearch for Anthropic provider
- Created anthropic.rs provider with native Anthropic API format
- Added web_search tool to Anthropic requests when enabled
- Anthropic uses /messages endpoint with x-api-key authentication
- Updated query_builder to route Anthropic to new provider
Implements websearch functionality for Anthropic Claude models.
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* feat(ai): implement websearch for Gemini/GoogleAI provider
- GoogleAI now uses completion endpoint (other.rs) for text instead of responses API
- Added Google Search grounding when websearch is enabled
- Uses google_search_retrieval tool in request when has_websearch is true
- Updated parse methods to use OtherQueryBuilder for completion endpoint
Implements websearch functionality for Google Gemini models.
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
* fix frontend
* fix anthropic and openai
* better for gemini
* structured output
* cleaning
* fix validate tool
* fixes
* cleaning
* cleaning
* fix for openai
* no responses api for azure
* fixes
* fix
* add tests for ai agent
* avoid panic
* better tests
* test user images
* fix tool choice
* always use streaming backend side
* big cleaning
* show annotations plus agent action for open ai websearch use
* show annotations plus agent action for anthropic websearch use
* show annotations plus agent action for google websearch use
* nit forntend
* rm
* fix
* add test for image ouptut
* fix for azure
* add in openflow
* fix
* fix
* nit tests
* fixes
---------
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
Co-authored-by: centdix <farhadg110@gmail.com>
* handle messages array for ai agent
* better
* nit
* make tool_calls and tool_call_id nullable
* fix empty json behavior
* nits
* cleaning
* feat(backend): replace messages/messages_context_length with history oneOf field
Replace the separate 'messages' array and 'messages_context_length' fields with
a single 'history' field that uses a oneOf discriminator.
The 'history' field can be either:
- 'auto' mode: automatically manages conversation history with memory, takes a
'context_length' number parameter
- 'manual' mode: bypasses memory and uses explicitly provided messages array
Backward compatibility is maintained: if 'messages_context_length' is provided
in the old schema format, it is automatically converted to 'auto' mode with the
specified context_length.
Co-authored-by: centdix <centdix@users.noreply.github.com>
* feat(frontend): replace messages/messages_context_length with history oneOf field
Replace the separate 'messages' array and 'messages_context_length' fields with
a single 'history' field in the AI agent schema.
The 'history' field uses a oneOf discriminator with two modes:
- 'auto': { mode: 'auto', context_length: number } - automatically manages
conversation history with memory
- 'manual': { mode: 'manual', messages: array } - bypasses memory and uses
explicitly provided messages
The schema includes comprehensive descriptions for each mode explaining the
behavior. The order array has been updated to include 'history' in place of
the old 'messages_context_length' and 'messages' fields.
Co-authored-by: centdix <centdix@users.noreply.github.com>
* fix(frontend): add support for 'mode' discriminator in oneOf rendering
Update ArgInput.svelte to properly handle oneOf schemas that use 'mode' as the
discriminator field, in addition to the existing 'kind' and 'label' support.
Changes:
- Updated tagKey derivation to check for 'mode' first, then 'kind', then 'label'
- Added 'mode' to the onOneOfChange function to track mode changes
- Added 'mode' to the list of keys excluded from enum validation
- Added 'mode' to hiddenArgs to prevent it from being shown in the form
- Added title fields to the history oneOf variants in flowInfers.ts
This allows the AI agent's history field to properly render with toggle buttons
for 'auto' and 'manual' modes.
Co-authored-by: centdix <centdix@users.noreply.github.com>
* fixes
* frontend fix
* nit
* cleaning
* cleaning
* better
* reword
* reword
---------
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: centdix <centdix@users.noreply.github.com>
* backend draft
* fix for tool and streaming
* do frontend side
* working
* working tools
* rm
* handle list endpoint
* handle for ai agents
* fix for models requiring inference id
* cleaning
* fix desc issue
* fix tool usage
* fix structured output
* cleaning
* fix for api
* rm
* fix input images
* cleaning
* chore: use aws sdk (#7156)
* feat(ai): Add AWS SDK dependencies for Bedrock integration
- Add aws-sdk-bedrockruntime v1.113.0
- Add aws-credential-types for bearer token authentication
- Update rustls to v0.23.35 for compatibility
- Dependencies added to windmill-common for AI features
* feat(ai): Add bearer token provider for Bedrock authentication
- Implement BearerTokenProvider using aws_credential_types
- Simple token-based auth using API keys from Windmill resources
- Add basic unit tests for provider creation
- Export bedrock_auth module in lib.rs
* feat(ai): Add Bedrock client wrapper with region extraction
- Implement BedrockClient wrapper around AWS SDK client
- Bearer token authentication integration
- Extract AWS region from Bedrock base URL automatically
- Comprehensive unit tests for region extraction
- Make aws-config non-optional dependency for AI features
- Update feature flags to reflect new dependency structure
* cargo
* feat(ai): Implement non-streaming Bedrock via AWS SDK
Use official AWS SDK instead of manual HTTP requests for better type safety and maintainability. Implements the Bedrock converse() API for non-streaming requests with proper bearer token authentication and message format conversion between OpenAI and Bedrock formats.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* refactor(ai): Eliminate Simple* conversion types for Bedrock SDK
- Move AI types to windmill-common/src/ai_types.rs for shared access
- Update bedrock_converters to work directly with OpenAI types
- Remove ~200 lines of conversion boilerplate from ai_executor.rs and bedrock.rs
- Remove unused imports to clean compilation warnings
- Benefits: 50% fewer conversion steps, no information loss, easier maintenance
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* feat(ai): Add streaming support for AWS Bedrock SDK
- Implement converse_stream() for Bedrock streaming responses
- Use EventReceiver.recv() to process stream events
- Extract text deltas using bedrock_stream_event_to_text()
- Send TokenDelta events to StreamEventProcessor for real-time updates
- Refactor request building to eliminate duplication between streaming and non-streaming
- Clean, minimal implementation following AWS SDK patterns
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* revert flake change
* fix
* feat(ai): Add tool calls and image support for Bedrock streaming
**Phase 1: Streaming Tool Call Support**
- Add stream event processing functions in bedrock_converters.rs:
- bedrock_stream_event_to_tool_start() - Extract tool use start from ContentBlockStart
- bedrock_stream_event_to_tool_delta() - Extract tool input deltas from ContentBlockDelta
- bedrock_stream_event_is_block_stop() - Detect ContentBlockStop events
- streaming_tool_calls_to_openai() - Convert accumulated tool calls to OpenAI format
- Update ai_executor.rs streaming loop with tool call accumulator (HashMap)
- Track current tool use ID during streaming
- Send ToolCallArguments events to StreamEventProcessor
- Return accumulated tool calls instead of empty vector
**Phase 2: Image Input Support**
- Add parse_image_data_url() to extract format and base64 data from data URLs
- Add content_part_to_block() to convert ContentPart to Bedrock ContentBlock
- Refactor convert_message() to handle multi-part content with images
- Support ImageUrl conversion to Bedrock ImageBlock with proper format (png/jpeg/gif/webp)
- Import AWS SDK image types: ImageBlock, ImageSource, ImageFormat
- Keep content_to_text() helper for system message text extraction
**Benefits**:
- ✅ Tool calling now works in both streaming and non-streaming modes
- ✅ Images are properly converted instead of being silently dropped
- ✅ Structured output works in streaming (uses tool calling)
- ✅ Full feature parity with manual HTTP implementation
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* cleaning
* fix(ai): Add S3 image support and structured output for Bedrock
**Fixes:**
1. **S3 Image Support**: Call prepare_messages_for_api() before Bedrock SDK path to convert S3Objects to ImageUrls
- Downloads images from S3 and encodes as base64 data URLs
- Ensures images are properly handled in both streaming and non-streaming modes
2. **Structured Output**: Add ToolChoice::Any when structured output tool is present
- Forces Bedrock to call the structured_output tool
- Ensures JSON schema compliance for structured output
- Works in both streaming and non-streaming modes
**Changes:**
- ai_executor.rs: Call prepare_messages_for_api() for Bedrock SDK path
- ai_executor.rs: Set tool_choice to Any when structured_output_tool_name is present
- aws_bedrock.rs: Remove unused ToolChoice imports (used via full path in worker)
**Testing:**
- ✅ S3 images are now downloaded and converted before API call
- ✅ Structured output now forces tool usage with ToolChoice::Any
- ✅ Both work in streaming and non-streaming modes
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* cleaning
* cleaning
* cleaning
* better error
* cleaning
* cleaning
* rm
* rename
* apply region
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix default
* no panic
* no print
* use utils file
* cleaning
---------
Co-authored-by: Claude <noreply@anthropic.com>
