AI Agentic Assistance

Nodeable includes an AI agentic assistance layer that blends branch context with policy-gated tool execution.

What this gives you

• Grounded context responses with source attribution
• Live branch graph snapshots for AI workflows
• Execution sessions that can be resumed, canceled, or rolled back
• Workspace AI memory for reusable constraints and standards
• Policy checks before tool execution
• Branch drift warnings before risky changes

Core workflows

1) Ask for grounded context

Use context query when you want evidence-backed answers from your current branch:

• POST /api/workspaces/:workspaceId/ai/context/query
• Optional controls include graph/wiki/branch metadata and result limits

You receive a grounded envelope with scored results and source handles.

2) Inspect live graph context

Use the live graph endpoint when a workflow needs direct node+edge snapshots:

• POST /api/workspaces/:workspaceId/ai/context/live-graph

3) Check branch drift risk

Before large write operations, check drift:

• GET /api/workspaces/:workspaceId/ai/branches/:branchId/drift

The response includes baseline branch info, lag signal, and risk level.

4) Run AI execution sessions

Create and manage execution sessions:

• POST /api/workspaces/:workspaceId/ai/executions
• GET /api/workspaces/:workspaceId/ai/executions/:executionId
• POST /api/workspaces/:workspaceId/ai/executions/:executionId/resume
• POST /api/workspaces/:workspaceId/ai/executions/:executionId/cancel
• POST /api/workspaces/:workspaceId/ai/executions/:executionId/rollback

Use these to persist multi-step AI work with explicit lifecycle state.

5) Manage workspace AI memory

Store durable conventions and constraints for your workspace:

• GET /api/workspaces/:workspaceId/ai/memory
• POST /api/workspaces/:workspaceId/ai/memory
• PATCH /api/workspaces/:workspaceId/ai/memory/:memoryId
• DELETE /api/workspaces/:workspaceId/ai/memory/:memoryId

Memory entries support draft/approved/archived workflows.

6) Keep AI chat on the same conversation thread

AI chat continuity depends on sending the same conversationId on each follow-up turn.

Chat endpoints:

• POST /api/workspaces/:workspaceId/ai/chat
• GET /api/workspaces/:workspaceId/ai/conversations
• GET /api/workspaces/:workspaceId/ai/conversations/:conversationId

How continuation works:

• On chat start, API returns the conversation ID in X-AI-Conversation-Id
• The first chat status SSE event also includes conversationId (operation: "chat")
• Web client stores this ID and sends it back on the next POST /ai/chat

Why this matters:

• Chat history (including assistant tool calls/results) is persisted per conversation
• Follow-up turns replay persisted history to the model
• If a turn is sent without the prior conversationId, a new conversation starts and prior tool usage context is not replayed

Policy and safety model

Tool calls are evaluated against capability policy before execution. Denied calls are blocked and logged.

Admin capability endpoints:

• GET /api/admin/ai/tool-capabilities
• PATCH /api/admin/ai/tool-capabilities/:capabilityId
• POST /api/admin/ai/tool-capabilities/:capabilityId/test-policy

Practical team usage

• Keep memory entries for naming conventions and architecture rules
• Use drift checks before broad branch updates
• Prefer execution sessions for multi-step changes instead of one-off prompts
• Keep higher-risk tools approval-gated for production workspaces

BYOK configuration (TEAM workspaces)

If your workspace uses BYOK mode, you can configure how many tool-calling steps AI chat may run before it must stop.

Where to set it:

1. Open workspace settings and go to AI Service Fleet
2. Switch provider mode to Bring your own key (BYOK)
3. Create or edit a BYOK provider
4. Set Max Agent Steps

Behavior:

• Valid range is 1 to 20
• Default is 8
• This affects the AI chat multi-step tool loop for that provider

If your workspace is in Managed mode, this value is controlled by your admin-managed provider configuration.