Code Confluence Changelog & Releases

Highlights

This release closes the loop on AGENTS.md by automatically opening a pull request with generated artifacts as part of the repository workflow. Repository ingestion is simplified to a single git URL, the CLI groups service lifecycle commands under one namespace, and the documentation site migrates to Vite+.

What's New

Automatic AGENTS.md Publishing

• Wired automatic AGENTS.md pull-request publishing into the repository workflow
• Batched artifact files into a single git tree commit

Simplified Repository Ingestion

• Sent only the git URL for repository ingest requests from the frontend
• Auto-derived repository identity from the git URL during ingestion

CLI and Docs

• Grouped service-related commands under service (run, destroy, stop, update) and added setup commands
• Fixed documentation and migrated the docs site to Vite+

Bug Fixes

• Preserved hostname:port in HTTPS git URL parsers
• Guarded the qualified-name parse inside the PR publish try block
• Pinned the agent prompt to the explicit AGENTS.md path
• Added the URLs required for probing the flow bridge
• Pinned the package version instead of using latest

Highlights

A small release that adds setup commands to streamline first-time configuration of a local Code Confluence environment.

What's New

Setup

• Added setup commands for getting a local environment configured

Highlights

This release scaffolds the new unoplat-code-confluence-cli, giving the project a first-class command-line entry point for running the stack. The CLI uses python-on-whales for its Compose runtime, models its runtime state with pydantic, and is wired into the build and release pipelines.

What's New

Command-Line Interface

• Scaffolded the initial unoplat-code-confluence-cli
• Used python-on-whales for the Compose runtime
• Migrated CLI runtime models from dataclasses to pydantic
• Added a dev dependency group with ruff and pyrefly
• Added the CLI package to the build and release pipelines

Bug Fixes

• Removed an unnecessary docker compose check and made the CLI release step reliable

Highlights

This release improves the application release pipeline, publishing a release manifest and Docker Compose asset on every app release and hardening the workflow to fail fast on validation errors.

What's New

Release Artifacts

• Published a release manifest and Docker Compose asset on app release

Bug Fixes

• Failed fast on release validation errors
• Pinned the release action and used a scoped GitHub token for security hardening

Highlights

This release replaces structural-signature extraction with BaseSourceContext detection primitives in the ingestion pipeline and routes query-engine output by workflow status, prefetching the prior run to decide how results are dispatched.

What's New

Detection Primitives

• Replaced structural-signature extraction with BaseSourceContext detection primitives in ingestion

Status-Based Output Dispatch

• Pre-fetched the previous workflow and used status-based output dispatch in the query engine

Bug Fixes

• Made GitHubUser fields optional and handled null display values in the frontend
• Fixed a logging issue

Highlights

This release unifies the AGENTS.md generation and refresh experience into a single Generate/Update action, backed by idempotency guards so repeated triggers stay safe. The query engine gains incremental dependency-guide delta fetching and a no-change carry-forward path so unchanged repositories skip redundant work.

What's New

Unified Agents.md Generation

• Replaced separate refresh and generate actions with a single Generate/Update Agents.md action in the frontend
• Added a post-refresh AGENTS.md update trigger with an idempotency guard in the flow bridge

Incremental Query Engine

• Added dependency-guide delta fetching with incremental cache reuse
• Added an engineering-workflow no-change sentinel and carry-forward to skip unchanged work
• Added an idempotency guard for the agent run endpoint and registered the fetch-delta activity
• Updated AgentMdOutput and added repository snapshot models

Bug Fixes

• Passed loguru structured log kwargs explicitly in the flow bridge
• Passed a typed envelope with trace context to AgentMdUpdateActivity
• Preserved the validation traceback in the query engine
• Used a Literal type for the EngineeringWorkflowNoChange sentinel and clarified prompts
• Renamed business_logic_domain to business_logic across all packages

Highlights

This patch release focuses on peak-memory reduction and runtime reliability, streaming database reads and file discovery instead of materializing them, eliminating redundant source parses in the detection pipeline, and stabilizing Temporal workflow timeouts.

Bug Fixes

Memory and Streaming

• Streamed SQLAlchemy result rows and used set[str] for library names in the backend
• Streamed framework features from PostgreSQL to reduce peak memory in the engine
• Streamed file discovery as an iterator in the parser to reduce peak memory
• Threaded pre-parsed source context through the detection pipeline to eliminate redundant parses

Runtime and Temporal

• Stabilized workflow timeouts and source decoding
• Added an env-var-overridable debug_timeout to all activity configs and the missing debug_timeouts module imported by the worker manager
• Restricted dev-workflow execute to help probes and used direct pydantic envelopes in Temporal runners
• Always overwrote markdown artifacts in the renderer, removing the stale-check read

Engine and Parser

• Used the pydantic-ai model API to render tool results in the event stream
• Extracted _iter_logical_lines to simplify pip requirements parsing
• Migrated framework feature repository tests to the streaming API and updated tests to the new source context API

Highlights

This release significantly trims the stack's memory footprint by moving basic workflow search operations off OpenSearch and onto PostgreSQL, and improves the agent events experience with reliable scroll restoration through long histories.

What's New

Leaner Workflow Search

• Moved basic workflow search operations from OpenSearch to PostgreSQL, reducing the memory footprint by roughly 700–900 MB

Agent Events

• Added anchor-based scroll restoration for older history via an imperative handle

Bug Fixes

• Always skip tool.result events when building display items
• Used Any for workflow interceptor return types to satisfy the type checker
• Used useLayoutEffect for scroll snap reset on codebase change

Highlights

This release reworks how agents are assembled and run, replacing bespoke file-reading tools with an audited console capability and swapping confidence scoring for real sandbox verification. The query engine migrates to the pydantic-ai Capabilities API with local backend rulesets, a new OpenMetadata connector lands, and first-class local dev tooling makes the stack easier to run.

What's New

Agent Assembly and Tooling

• Replaced custom repo-inspection and read_file_content tools with an audited console capability across the business-domain and assembly agents
• Dropped the DuckDuckGo tool in favor of a builtin web-fetch capability and extended provider support
• Extracted Docker lifecycle, path helpers, and resolver logic into a dedicated agent backend, integrated via AgentBackendLifecycle in the worker and workflows

Sandbox-Verified Development Workflow

• Replaced confidence scoring in the development workflow with sandbox verification of changes
• Added a unified thinking-level parameter with a high default across all model providers

Query Engine

• Migrated to the pydantic-ai Capabilities API with local backend rulesets and upgraded pydantic-ai to v1.83
• Added a markdown validation and rendering pipeline plus dependency-guide and app-interfaces write activities
• Extended framework feature validation models and repository support
• Removed the agents_md_updater agent and the section-updater pattern

Connectors and Local Dev

• Added a Code Confluence → OpenMetadata connector package with optional snapshot lookup
• Added a Tiltfile, docker-compose, and dev Dockerfiles for a smoother local setup
• Virtualized agent group events with sticky auto-scroll and updated frontend schema/utils for the new agent output structure

Bug Fixes

• Added authenticated git clone support for private GitHub repositories
• Fixed Node.js symlink dereferencing in the query-engine Docker image
• Updated query-engine agent prompts, runners, config, and Dockerfile
• Removed stale command_discovery_requirements and examples from the workflow prompt
• Removed the obsolete litellm TypeScript framework definition

Highlights

This release introduces the v4 capability-operation detection model, expands framework and MCP coverage, adopts structured feature identity across the platform, and hardens the docs and ingestion pipelines to match the new hierarchy.

What's New

Capability-Operation Schema v4

• Introduced the v4 detection schema with a capability-operation-detector hierarchy
• Migrated framework definitions and schema contracts to the new structured hierarchy
• Replaced flat framework feature identifiers with canonical capability_key and operation_key fields

Expanded Platform Coverage

• Added FastMCP client support with mcp_client capability and FastMCP toolset client definitions
• Added Firebase TypeScript framework detection
• Expanded canonical capability families and renamed capability identifiers to match the new schema model

Platform-Wide Structured Feature Identity

• Adopted structured framework feature identity across commons, ingestion, and query engine components
• Flattened capability-operation pairs into composite feature keys during ingestion where needed for downstream compatibility
• Added deterministic commit SHA pinning for network clone tests

Docs and Agent Assembly

• Updated the docs route and framework feature catalog to render the new capability-operation hierarchy
• Added hideDescription support and improved DIDS link rendering in the docs UI
• Added a Temporal agent_assembly package with catalog, assembler, and agent definitions

Bug Fixes

• Raised the agent confidence threshold to 0.70
• Guarded the docs catalog against null feature entries
• Fixed docs project typecheck and build failures
• Split hash fragments into a dedicated prop for TanStack Router links
• Aligned ingestion definitions, query engine mappings, tests, and schema names with canonical capability keys
• Improved ingestion recovery during git pull merge conflicts
• Upgraded commons dependencies to v0.47.0 across ingestion and query engine services

Highlights

This patch release substantially hardens detection and ingestion by introducing structured codebase detection models, improving TypeScript monorepo support, and refactoring backend services into clearer modules while also polishing alpha UI messaging and CI behavior.

Bug Fixes

Codebase Detection

• Added structured detection data models for rules, evidence, results, and workspace metadata
• Added TypeScript monorepo detection and reorganized detection logic into clearer language-specific namespaces
• Extracted shared parser infrastructure and modernized structured logging types
• Propagated workspace metadata through query engine services, models, and tests
• Removed legacy flat-layout files superseded by the namespace reorganization

Backend and Framework Support

• Added APIRouter support to FastAPI framework definitions
• Improved TypeScript workspace detection for monorepos
• Added pnpm-workspace.yaml support and negation glob handling
• Extracted repository, operations, credential, provider, and flag routers plus shared utilities from main.py
• Relaxed GitHub helper language assertions to better tolerate valid repository states

Reliability and UX

• Scoped PR concurrency by head repository and refreshed nested uv lockfiles explicitly in CI
• Added vendored turbo monorepo fixture files to strengthen test coverage
• Updated the alpha status treatment near the app name for clearer UI signaling

Highlights

This release upgrades the framework catalog experience with a richer table, better filtering and pagination, and a smoother contribution workflow, while tightening synchronization between docs data and generated artifacts.

What's New

Framework Catalog Experience

• Added a framework catalog table that makes supported frameworks and capabilities easier to browse
• Introduced filtering and pagination so larger catalogs remain navigable as coverage grows

Contribution Workflow

• Added a GitHub issue template to make framework requests and contribution feedback more structured

Bug Fixes

• Fixed duplicated information across generated artifacts and AGENTS.md
• Corrected numeric filter synchronization so parent state updates propagate properly
• Ensured the prebuild step generates catalog JSON consistently
• Refreshed Python uv lockfiles before CI lock validation runs

Highlights

This release refreshes the docs experience with updated functionality coverage, new screenshots, and clearer release documentation while trimming unnecessary preview guidance.

What's New

Documentation Refresh

• Updated the docs to reflect newly shipped functionality across the product
• Added fresh screenshots to better illustrate the current user experience
• Expanded release documentation so the changelog stays aligned with the application

Bug Fixes

• Removed preview output that was no longer needed
• Eliminated duplicate troubleshooting guidance to keep the docs cleaner and easier to scan

Highlights

This release introduces agent workflow cancellation from the Operations UI, managed AGENTS.md context blocks with git ref resolution, duplicate ingestion prevention, and a fully redesigned agent feedback experience with binary per-agent ratings.

What's New

Agent Workflow Cancellation

• Cancel in-flight agent generation directly from the Operations UI
• Full backend cancellation propagation ensures clean teardown of running workflows

Managed AGENTS.md Context Block

• Git ref resolution and managed block bootstrap for structured AGENTS.md updates
• Ensures consistent, version-aware context blocks across generated documentation

Duplicate Ingestion Prevention

• 409 guard prevents submitting the same repository for ingestion multiple times
• Clear toast notification when a duplicate ingestion is attempted

Redesigned Agent Feedback

• Replaced the emoji-based rating sheet with a streamlined thumbs up/down dialog
• Binary per-agent ratings for all four agent types: Development Workflow Guide, Dependency Guide, Business Domain Guide, and App Interfaces Guide
• Enhanced preview with tech stack badges and "Copy All" functionality
• Per-codebase sections with progress badges for tracking rating completion

Bug Fixes

• Fixed toast notifications for duplicate ingestion requests
• CI lockfile refresh for Python packages

Highlights

This release adds AWS Bedrock as a model provider, introduces a command palette for faster navigation, enables in-app feedback, improves the TypeScript beta with stronger detection capabilities, adds LLM-based DIDS validation for low-confidence matches, and reworks catalog navigation to be language-first.

What's New

AWS Bedrock Support

• Added AWS Bedrock as a model provider with enhanced Anthropic model cache support
• Broadens deployment options for teams running on AWS infrastructure

Command Palette

• Introduced a command palette for easier navigation and quick access to actions across the app

In-App Feedback

• Added a feedback dialog accessible from the command palette with a dedicated backend endpoint for collecting user feedback

Language-First DIDS Catalog

• Reworked the catalog hierarchy to use a language-first navigation model for more intuitive browsing of ingested codebases

TypeScript Beta Improvements

• Stronger grammar detection and Next.js endpoint detection for more accurate TypeScript codebase analysis

LLM-Based DIDS Validation

• Added an LLM-based validator for low-confidence DIDS matches, improving ingestion accuracy with confidence metadata schema for ingestion contracts

Additional Framework Support

• Extended Python and TypeScript framework detection coverage

Documentation Updates

• Added roadmap for skills and blueprint generation

Bug Fixes

• Agent events now derive live status from the latest event for accurate real-time state
• Overhauled events view for more intuitive browsing with support for up to 100k tool results
• Improved real-time sync performance with virtual list and 3-phase schema (ElectricSQL)
• Aligned tool call/result UI colors to the design system
• Responsive dialog sizing via CVA variants
• Fixed codebase dropdown visibility and adaptive sizing
• Fixed command palette positioning
• Various schema, Docker, and dependency fixes

Highlights

This release delivers a comprehensive SEO overhaul for the documentation site, redesigns the changelog into a single-page timeline, and aligns the frontend with design-system tokens and primitives.

What's New

Comprehensive SEO

• Added meta tags, canonical links, JSON-LD structured data, and a robots.txt for improved search engine discoverability
• SITE_URL is now sourced from import.meta.env for correct canonical URL generation across environments

Changelog Timeline Redesign

• Redesigned the changelog into a single-page Commit-theme timeline for a more cohesive reading experience
• Added v0.28.0 changelog entry

Design-System Alignment

• Aligned design-system tokens and primitives across the frontend for visual consistency

Bug Fixes

• Fixed design-system token and primitive alignment issues
• Fixed Read tool call pagination with clamp/no-op correction
• Fixed SITE_URL to be sourced from import.meta.env instead of hardcoded values

Highlights

This release introduces AGENTS.md PR creation, giving you the ability to create pull requests that keep your AI agent guides in sync with the codebase. It also adds tool call and result pairing in agent events for better observability, and build-time image optimization for faster page loads.

What's New

AGENTS.md PR Creation

A complete workflow for creating pull requests to update AGENTS.md files:

• New updater agent that generates AGENTS.md content and creates PRs via one-shot endpoints
• One-shot PR creation endpoints with PrMetadata model for structured PR metadata
• Frontend UI for triggering and monitoring AGENTS.md PR creation workflows
• One-shot semantics ensure the entire generate-and-PR flow runs as a single atomic operation

Tool Call and Result Pairing in Agent Events

• Agent event streams now pair each tool call with its corresponding result, making it easier to trace and debug agent workflows in real time

Performance: Build-time Image Optimization

• Images are now optimized at build time with lazy loading, reducing initial page load size and improving Core Web Vitals

Documentation Branding

• Added brand logo to the navigation header and updated 404 page branding for a consistent visual identity

Bug Fixes

• Fixed changelog sorting to use version number instead of release date for correct ordering
• Fixed broken Discord links across docs, app, and README
• Fixed sidebar scroll issue caused by dots in banner ID creating invalid CSS selectors
• Fixed incorrect roadmap card links
• Aligned commons version dependencies across packages

Highlights

A major feature release introducing Codex as a new provider via OAuth PKCE authentication, built-in web search support for capable providers, smarter dependency filtering, and significant improvements to the engineering workflow agent. This release also refreshes brand assets across the platform.

What's New

Codex OAuth PKCE Integration

Added Codex as a new model provider with a full OAuth PKCE authentication flow:

• Secure browser-based OAuth flow with PKCE for Codex provider
• Seamless callback handling and token management
• Full integration with the existing model configuration system

Built-in Web Search

• Providers with native web search capability can now use it directly within agent workflows, enabling real-time information retrieval during code analysis

Dependency Filtering

• Restricted dependency discovery to runtime-only dependencies, filtering out development and test dependencies for cleaner and more relevant dependency guides

Engineering Workflow Agent Improvements

• Simplified output schema for clearer and more predictable results
• Improved system prompts for better agent accuracy
• Enhanced event stream debugging and workflow completion observability
• Redesigned frontend UI for engineering workflow operations

Delete Model Configuration

• Added a UI control to delete active model provider configurations, giving users full lifecycle management of their provider setups

Documentation Updates

• Updated project documentation to reflect new features and workflows

Bug Fixes

• Simplified agent names for better readability across the UI
• Refreshed brand assets with new logo and visual identity
• Reverted binding address to 127.0.0.1 for local development stability
• Cleaned up code smells in the model factory
• Enhanced error detail logging for model provider issues
• Replaced loguru with workflow.logger in the workflow outbound interceptor for consistency
• Removed duplicate object property in ModelConfigForm
• Updated sidebar logo imports and added "Made with love from India" footer
• Fixed networking issues for the OAuth callback listener
• Preview results now display correctly during partial errors
• Increased axios client timeout from 10s to 120s to prevent premature request failures

Highlights

Building on the backend interface detection introduced in v0.25.0, this release brings full frontend support for viewing and exploring application interfaces directly in the UI.

What's New

Application Interface Views

A new frontend experience for exploring detected interfaces across your codebase:

• Inbound interfaces — View all entry points into your application, including REST APIs, gRPC handlers, and event listeners
• Outbound interfaces — Explore all external calls your application makes, such as HTTP requests, database queries, and message publishing
• Internal interfaces — Discover internal module boundaries and inter-service communication patterns within your codebase

Highlights

This release introduces automatic detection of inbound and outbound interfaces starting with Python in codebases, giving developers clearer visibility into how their code interacts with external systems. It also adds a dependency guide banner to the frontend.

What's New

Inbound and Outbound Interface Detection

Capable of automatically identifying and classifying interfaces in your codebase:

• Detects inbound interfaces (APIs, endpoints, event handlers) that receive external requests
• Detects outbound interfaces (HTTP clients, database calls, message producers) that communicate with external systems

Dependency Guide Banner

• Added a prominent banner highlighting the Dependency Guide feature, making it easier for users to discover and use this capability

Bug Fixes

• Quick start improvements — Enhanced the quick start documentation for a smoother onboarding experience

Highlights

This patch improves reliability for tool operations and enhances the quality of code descriptions generated for open source models.

Bug Fixes

• Increased tool retries — Raised the retry count for tool operations to improve resilience when interacting with external services
• Improved Griffe-based descriptions — Enhanced the Griffe-based description generation to produce more accurate output, reducing errors when used with open source models

Highlights

A patch release addressing a multiprocessor conflict in the flow bridge service.

Bug Fixes

• Flow bridge multiprocessor conflict — Resolved a conflict that occurred when multiple processors were running concurrently in the flow bridge, ensuring stable operation under load

Highlights

A quick patch release to resolve a Docker build trigger issue in the CI/CD pipeline.

Bug Fixes

• Docker CI/CD trigger fix — Resolved an error in the Docker build trigger by pushing logging changes that corrected the build pipeline configuration

Highlights

This release introduces the Dependency Guide Agent, providing coding agents with contextual knowledge about which dependencies to use for specific tasks. Additionally, comprehensive MCP tool management capabilities have been added.

What's New

Dependency Guide Agent (Alpha)

A new agent that generates repo-specific guides explaining what each dependency is for and when to use it, helping coding agents make informed decisions about library usage.

• Frontend integration for dependency guide generation
• Backend workflow with pydantic fastmcp client for MCP servers
• Migration from previous agent with context (now using Exa MCP tool)

MCP Tool Management

• Full CRUD operations for MCP tools via the frontend
• Schema-based authentication support for MCP tools
• Dedicated MCP tool section for agent MD operations

UI Improvements

• Enhanced sidebar logo with improved visual presentation

Bug Fixes

• Safe logging utility showing only first/last characters of sensitive information
• Intuitive error display for agent MD operations
• DateTime offset fixes across backend operations
• Migration from standard FastAPI MCP client to pydantic fastmcp for async reliability issues while doing mcp operations in temporal activities
• MDX editor styling fixes removing unnecessary class applications
• Test container v4 API compatibility fixes for query engine tests
• Tool configuration timestamp offset corrections

Highlights

This release focuses on improving the documentation site with an upgraded roadmap and enhanced release notes presentation.

What's New

Documentation Improvements

• Upgraded roadmap — Refreshed the project roadmap with clearer milestones and updated priorities
• Enhanced release notes — Improved the release notes format and presentation for better readability and discoverability

Highlights

This release marks a significant architectural simplification by migrating from Neo4j to PostgreSQL for knowledge graph storage, reducing operational complexity while maintaining performance through optimized rdbms schema.

What's New

PostgreSQL Migration (Major)

• Migrated ingestion and query engine services to PostgreSQL
• Removed Neo4j related models, schema, and dependencies

Infrastructure Savings

By removing Neo4j and consolidating on PostgreSQL (which was already required for Temporal workflow data), self-hosted deployments benefit from reduced resource requirements:

Documentation Improvements

• Added common links in sidebar for better navigation
• Zoom and preview support across all MDX docs
• Submit feedback documentation added
• Changelog RSS feed with theme alignment

Bug Fixes

• System prompt refinements for improved agent descriptions
• Validations and Retries across agent harness to improve compatibility with open source models.
• Framework/lib features' detection across codebase for inbound and outbound interfaces refactored from structural signatures to source code analysis which will improve scalability across programming languages.
• Owner name fallback for user repositories
• Theme-based icon color adjustments
• Changelog sorting by release date and version number
• Documentation content improvements

Highlights

This patch release focuses on Docker improvements and codebase cleanup.

What's New

Query Engine Dockerfile Enhancement

Added the eza tool to the query engine Dockerfile for improved file listing and exploration capabilities during debugging and development.

Maintenance

• Removed legacy commented code to improve codebase maintainability
• Cleaned up unused code sections across the query engine

Highlights

This release brings important stability improvements through dependency pinning, enhanced documentation, and refined schema validation.

What's New

Documentation Updates

• Updated roadmap documentation to reflect current priorities
• Improved clarity on upcoming features and timelines

Schema Validation Improvements

Refined schema validation to ensure more robust data handling across the pipeline:

• Stricter type checking for ingested code structures
• Better error messages for validation failures

Bug Fixes

• Corrected test failure scenarios that were causing false negatives
• Fixed edge cases in the testing infrastructure

Highlights

This is a major feature release introducing auditable agent workflows, comprehensive feedback mechanisms, and significant UI improvements for real-time progress tracking.

What's New

Auditable Agent Workflows

Implemented full auditability for agent operations:

• Track every step of agent execution
• View detailed logs and state transitions
• Debug and analyze agent behavior with complete visibility

Integrated Feedback Mechanisms

Added feedback capabilities across all operations:

• Provide feedback on code analysis results
• Rate the quality of generated documentation
• Submit suggestions for improvement directly from the UI

GitHub Issue Submission

New error reporting capability that integrates directly with GitHub:

• One-click issue creation for encountered errors
• Automatic inclusion of relevant context and logs
• Streamlined communication between users and maintainers

Enhanced Form Infrastructure

Built reusable form components for better UX:

• Consistent form styling across the application
• Improved validation and error messaging
• Accessible form controls following WCAG guidelines

Real-time Progress Tracking UI

Improved the progress tracking experience:

• Live updates on workflow execution status
• Visual indicators for each processing stage
• Estimated time remaining for long-running operations

Highlights

This patch release addresses several important stability issues across the frontend and backend, including a major migration from Zustand to TanStack Query for provider state management and fixes for Neo4j graph merge conflicts.

Bug Fixes

• Provider state management overhaul — Migrated from Zustand store to custom hooks based on TanStack Query for managing provider server state across all related sections, replacing hardcoded child paths with dynamic TanStack Router params
• Shared axios client extraction — Extracted axios clients to a shared module and reduced stale time and fetch-on-mount behavior for the job status dialog
• Neo4j merge conflict fix — Updated CodeConfluenceGitRepository node creation to avoid Cypher merge conflicts in the graph database
• Provider-based user details — User details are now fetched based on the active repository provider state, improving reliability
• GitHub Enterprise and data residency — Updated support for GitHub Enterprise Server and data residency configurations