AtlasClaw v0.9.11 Released | Capability Routing, Knowledge Base Retrieval, and Runtime Stability Updates
Published: 2026-06-05 12:00
Introduction
v0.9.11 focuses on solving the enterprise challenge where “a single integration type maps to multiple actual systems” — for example, the same SmartCMP Provider connecting to both a development CMP and a production CMP, or the same Markdown Vault Provider connecting to multiple independent knowledge bases. For teams that have already integrated multiple business systems or document repositories, this release delivers value in three key areas: more accurate capability selection, clearer system instance boundaries, and more stable interaction execution.
Update Summary
• Provider Skills now support routing to specific Provider instances. A single Provider type can map to multiple system instances — for example, a development CMP and a production CMP, or multiple independent Markdown knowledge bases.
• Multi-step workflows are now judged by the model based on context and tool evidence to determine whether to continue the current capability, reducing dependence on fixed confirmation phrases.
• Markdown Vault has been rearchitected to retrieve Markdown files directly at runtime, eliminating database index maintenance overhead.
• Log rotation, MySQL connections, local model compatibility, and historical context handling continue to be refined.
Core Capability Routing and Runtime Boundaries
Core-side updates focus on capability selection, Provider instance isolation, historical context handling, and production deployment configuration. The key change is the separation of Provider Type from Provider Instance: a Provider Type represents an integration capability, while a Provider Instance represents a specific system or data source.
This solves the target selection problem when connecting multiple systems. For example with SmartCMP, a team can simultaneously configure a development CMP and a production CMP; with Markdown Vault, a team can simultaneously configure a product knowledge base, an operations Runbook knowledge base, and a customer project knowledge base.
1. Provider Skill Instance-Level Routing
AtlasClaw now organizes Provider Skills using instance-scoped capability identifiers. When a user selects a capability via Slash Command, or an external system pre-selects a capability via Webhook, the runtime retains both the Provider Instance and Skill information.
The same Provider Type can be configured with multiple instances, with instance boundaries preserved during capability selection, tool projection, and execution phases.
The capability list hides internal Skill snapshots, presenting curated selectable capabilities to users and automation entry points.
2. Multi-Step Workflow Routing Judgment
Previously, some workflows relied on fixed confirmation phrases to determine whether to continue execution. v0.9.11 shifts this to model-based judgment, combining the conversation, selected capabilities, and tool evidence to decide whether to continue the current workflow, re-select a capability, or directly generate a response.
3. Historical Context and Model Compatibility
This update fixes several details in history replay and model adaptation, including preserving the user’s original expressions, preventing runtime system prompts from being duplicated into conversation history, and normalizing system prompts for OpenAI Chat API compatible models.
• Multi-turn conversations are less affected by stale runtime prompts.
• User’s original expressions are more fully preserved during error recovery and evidence tracing.
• System prompt compatibility is more stable for local or private models such as vLLM and Qwen.
4. Deployment and Operations Configuration
Production-related configurations continue to be expanded, including container log rotation, systemd/logrotate templates, and MySQL connection pool pre-ping configuration.
• Container logs rotate according to standard policies, reducing the risk of unlimited log growth during long-running operations.
• VM deployments can reference systemd and logrotate templates for service management and log retention.
• MySQL connection keep-alive strategies are configurable to adapt to different network and connection pool environments.
Provider Updates: Markdown Vault and SmartCMP
Provider-side updates primarily involve Markdown Vault and SmartCMP. Markdown Vault reduces knowledge base onboarding and maintenance costs, while SmartCMP adds resource request field contracts to reduce field assembly errors in automated requests.
1. Markdown Vault Multi-Knowledge-Base Direct Query
The Markdown Vault Provider has switched from a database indexing model to runtime scanning of Markdown files. After configuring the vault path, include/exclude rules, and context budget, internal documents, Runbooks, design records, or Obsidian notes can be connected to AtlasClaw as read-only knowledge bases.
• SQLite/MySQL index tables are no longer required, and no separate index refresh step is needed.
• Retrieval supports model-expanded keywords and returns evidence snippets constrained by the context budget.
2. SmartCMP Request Field Contracts
The SmartCMP Provider has added field rules for Compute/VM requests. Taking systemDisk as an example, this field must be submitted as an object — for example, a JSON object containing disk size; it cannot be a number or string, nor can it be moved under params.
• Fields for system disk, specifications, images, networks, and security groups are now closer to SmartCMP API requirements.
• Field names declared in generated Markdown must be preserved as-is, reducing request failures caused by field name rewriting.
Community and Resources
GitHub Main Repository: https://github.com/CloudChef/atlasclaw
Provider Plugin Repository: https://github.com/CloudChef/atlasclaw-providers
Project Website: https://atlasclaw.ai
Contact Email: atlasclaw@cloudchef.io
Contributions are welcome through Issues, Pull Requests, or Provider plugin contributions.
About Cloudchef
Cloudchef is a leading provider of cloud management and digital operations solutions, dedicated to helping enterprises accelerate their digital transformation through innovative cloud management platforms and services. With core products including the SmartCMP cloud management platform and AtlasClaw AI Agent, Cloudchef serves over 2,500 enterprise customers across finance, government, education, manufacturing, healthcare, and other industries. The company has been recognized as a Shanghai High-Tech Enterprise and has received multiple industry awards for its contributions to cloud management and IT operations innovation.