Feature Request: Context Window / Session Insights view (like OpenCode desktop)

[gaborho]

Feature Request

Version: 0.18.0 (Tauri)
OS: macOS Tahoe 26.5.2

Summary

Introduce a Context Window / Session Insights view in CodeNomad, similar to the one available in the OpenCode desktop app. This view provides detailed insights about the current session and is helpful for troubleshooting and understanding session behavior, usage, and cost.

Proposed Behavior

The view should be opened when the "circle" context icon (displayed above the session window) is clicked. It would surface session-level metadata and usage statistics in a single panel.

Fields to Display

• Session
• Provider
• Usage
• Output Tokens
• Cache Tokens
• Assistant Messages
• Sessions created
• Messages
• Model
• Total Tokens
• Input Tokens
• Reasoning Tokens
• User Messages
• Total Cost
• Last Activity

Motivation

This view is helpful for troubleshooting the session and gaining insights into token usage, cost, and activity, making it easier to understand and debug what is happening within a session.

Notes

A screenshot from the OpenCode desktop view can be attached/added to this issue later for reference. See here:

[pascalandr]

@gaborho Thanks for the suggestion. I'm not fully convinced that a separate Session Insights view would add enough value in its current form, because most of this information is already available in CodeNomad:

• context usage is displayed in the session header
• token usage and cost are available in the Status panel
• provider and model are visible through the model selector and message metadata
• detailed input, output, reasoning, cache, and cost metrics can be displayed for each assistant response
• session activity is already represented in the session list

The main difference would therefore be consolidating existing information into another panel. That could improve discoverability, but it would also duplicate UI and introduce ambiguity between metrics for the current model call, the current session, and the entire session thread including subagents.

A smaller improvement might be to make the existing context indicator clickable and use it to reveal or focus the current context and usage information in the Status panel, rather than creating a new dashboard.

Before implementing a larger view, could you clarify which troubleshooting workflow is currently difficult in CodeNomad and which specific information cannot be obtained from the existing UI? That would help determine whether we need a new view or only better access to the existing metrics.

[gaborho]

@pascalandr you are right, most of the information is already available. One example for troubleshooting: we access models through LiteLLM and Bedrock, and in case my budget is expired, I just see an error, but no details about the request. In such situation it would be great if I could see the raw messages with all the payloads. The context breakdown bar is also useful to discover what filled the context.

[pascalandr]

@gaborho Thanks, that use case makes the request clearer. CodeNomad already has an Instance Info view, accessible through the small information button at the top of the left panel, or from the command palette. It includes OpenCode process information, service status, environment details, and live raw server stdout/stderr logs.

Those logs may already expose additional LiteLLM or Bedrock errors, depending on what OpenCode writes to its process output. However, they are different from the raw structured session messages and provider request payloads you are describing.

Rather than introducing a separate Session Insights dashboard, extending the existing Instance Info view may be a better fit. It could gain a session-specific diagnostics section containing:

• active session, model and provider
• context usage and breakdown
• token and cost totals
• raw OpenCode session messages and parts

The context indicator could then open that diagnostics section directly.

One important distinction is that CodeNomad can expose the OpenCode session data it receives, but it may not have access to the exact HTTP request, headers, or response exchanged between OpenCode and LiteLLM/Bedrock. Raw diagnostic data may also contain sensitive prompts, source code, and tool output, so it would need to be explicitly revealed and handled carefully.

Could you check whether the existing Instance Info server logs contain the LiteLLM/Bedrock error details you need? That would help determine whether this is mainly a discoverability improvement or whether an additional raw session-data inspector is required.

[gaborho]

@pascalandr thanks for drawing my attention to the Instance Info view. Yes, there the issue is clearly visible. (I redacted some IDs.)

timestamp=2026-07-16T10:36:41.180Z level=INFO run=<RUN_ID> message=loop session.id=<SESSION_ID> step=0
timestamp=2026-07-16T10:36:41.190Z level=INFO run=<RUN_ID> message=stream providerID=genai-api-gw modelID=claude-haiku-4-5 session.id=<SESSION_ID> small=true agent=title mode=primary
timestamp=2026-07-16T10:36:41.192Z level=INFO run=<RUN_ID> message="llm runtime selected" llm.runtime=ai-sdk llm.provider=genai-api-gw llm.model=claude-haiku-4-5
timestamp=2026-07-16T10:36:41.208Z level=INFO run=<RUN_ID> message=process session.id=<SESSION_ID> messageID=<MESSAGE_ID>
timestamp=2026-07-16T10:36:41.209Z level=INFO run=<RUN_ID> message=stream providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary
timestamp=2026-07-16T10:36:41.216Z level=INFO run=<RUN_ID> message="llm runtime selected" llm.runtime=ai-sdk llm.provider=genai-api-gw llm.model=claude-sonnet-5
timestamp=2026-07-16T10:36:41.522Z level=ERROR run=<RUN_ID> message="stream error" providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary error.error="AI_APICallError: Budget has been exceeded! Current cost: 400.0053308000001, Max budget: 400.0"
timestamp=2026-07-16T10:36:43.526Z level=INFO run=<RUN_ID> message=stream providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary
timestamp=2026-07-16T10:36:43.536Z level=INFO run=<RUN_ID> message="llm runtime selected" llm.runtime=ai-sdk llm.provider=genai-api-gw llm.model=claude-sonnet-5
timestamp=2026-07-16T10:36:44.148Z level=ERROR run=<RUN_ID> message="stream error" providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary error.error="AI_APICallError: Budget has been exceeded! Current cost: 400.0053308000001, Max budget: 400.0"
timestamp=2026-07-16T10:36:47.918Z level=ERROR run=<RUN_ID> message="stream error" providerID=genai-api-gw modelID=claude-haiku-4-5 session.id=<SESSION_ID> small=true agent=title mode=primary error.error="AI_RetryError: Failed after 3 attempts. Last error: Budget has been exceeded! Current cost: 400.0053308000001, Max budget: 400.0"
timestamp=2026-07-16T10:36:48.158Z level=INFO run=<RUN_ID> message=stream providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary
timestamp=2026-07-16T10:36:48.176Z level=INFO run=<RUN_ID> message="llm runtime selected" llm.runtime=ai-sdk llm.provider=genai-api-gw llm.model=claude-sonnet-5
timestamp=2026-07-16T10:36:48.556Z level=ERROR run=<RUN_ID> message="stream error" providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary error.error="AI_APICallError: Budget has been exceeded! Current cost: 400.0053308000001, Max budget: 400.0"
timestamp=2026-07-16T10:36:56.574Z level=INFO run=<RUN_ID> message=stream providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary
timestamp=2026-07-16T10:36:56.587Z level=INFO run=<RUN_ID> message="llm runtime selected" llm.runtime=ai-sdk llm.provider=genai-api-gw llm.model=claude-sonnet-5
timestamp=2026-07-16T10:36:56.946Z level=ERROR run=<RUN_ID> message="stream error" providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary error.error="AI_APICallError: Budget has been exceeded! Current cost: 400.0053308000001, Max budget: 400.0"
timestamp=2026-07-16T10:37:12.951Z level=INFO run=<RUN_ID> message=stream providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary
timestamp=2026-07-16T10:37:12.970Z level=INFO run=<RUN_ID> message="llm runtime selected" llm.runtime=ai-sdk llm.provider=genai-api-gw llm.model=claude-sonnet-5
timestamp=2026-07-16T10:37:13.535Z level=ERROR run=<RUN_ID> message="stream error" providerID=genai-api-gw modelID=claude-sonnet-5 session.id=<SESSION_ID> small=false agent=build mode=primary error.error="AI_APICallError: Budget has been exceeded! Current cost: 400.0053308000001, Max budget: 400.0"

I noticed that even a small red popup is displayed when such error happens. Ok, at the moment no urgent need then from my side to have more implementation, however I like and support the idea of raw session-data inspector view (separate than Instance Info view).
Thank you if you consider to implement something similar like what OpenCode Desktop has.