Hi Studio team,
The licensing support team asked us to open a separate question about these ABP Studio AI issues.
Environment: Windows, ABP Studio 3.1.4, ABP application packages 10.7.0.
During planning and implementation, we repeatedly received:
“The conversation is too large for the selected model's context window. Please start a new chat or select a model with a larger context window.”
We also received:
“AI response timed out. The model took too long to respond. Please try again.”
“Unable to read data from the transport connection: An existing connection was forcibly closed by the remote host.”
Typical sequence:
- A session accumulates source reads, plan content and diagnostics.
- A subsequent request fails with one of these errors.
- Changing models does not reliably recover the existing session.
- A fresh session needs to inspect saved work to determine what completed.
- The fresh session can report no active plan even though the plan file exists, making continuation harder.
No underlying exception stack was displayed.
Your AI Agent Configuration documentation describes a default 300K context limit and Max, and says Max also increases reasoning effort: https://abp.io/docs/latest/studio/ai-agent-configuration
Could you please clarify:
Where exactly are these settings in Studio 3.1.4? What does 300K versus Max include-history, tool results, system instructions, reasoning and reserved output? What happens if the selected model supports less than 300K?
What is the recommended setting and workflow to avoid context errors without unnecessarily increasing credit consumption?
Does Studio automatically compact history or preserve continuation checkpoints? How should a new session reconnect an existing saved plan and continue only unfinished steps?
After a timeout or transport failure, how can we reliably determine whether a tool operation or file save completed before retrying?
Are context-limit rejections, timeouts, transport failures and automatic retries charged? Can you review our affected requests and restore or refund eligible credits?
How can we export session/request IDs, timestamps and usage records safely for your investigation? Please provide the exact steps and a private submission route if needed.
We are not assuming every failed request was charged or that all work was lost. We need an accounting review and a supported recovery workflow that avoids repeatedly paying to reconstruct the same context.
Please have the Studio engineering and billing teams review this. Thank you.
2 Answer(s)
-
0
- Licensing / billing / refunds / credit review (your items 5–6): please contact ABP via https://abp.io/contact (this is the supported/private route for billing review and account-specific investigations, including credit adjustments and sharing identifiers/logs).
- Technical behavior (your items 1–4): below is what’s verifiable from the current ABP Studio “AI Agent Configuration” documentation and generally supported Studio workflows.
1) Where are the “300K” vs “Max” settings in Studio 3.1.4, and what do they include?
- In ABP Studio, this is the AI Agent Configuration → Context Limit setting (the docs describe it as “controls the maximum context size used for agent runs”).
- 300K (default): Studio targets a maximum context size of ~300K for the agent run context.
- Max: “delegates to the model’s maximum context behavior.” So Studio will attempt to use whatever the selected model supports as its max context.
What the docs explicitly confirm:
- The limit applies to the agent run context (not just “chat messages”), and larger context increases latency and model cost.
- ABP recommends narrow AI scopes and targeted attachments to avoid large-context use when not needed.
What is not specified in the verified docs:
- A precise breakdown of whether the “context” number includes system instructions vs conversation history vs tool results vs reserved output vs internal reasoning tokens, or what Studio does if a chosen model supports less than 300K. (No verified public breakdown for this exact accounting.)
2) Recommended setting & workflow to avoid context errors without unnecessary credit usage
Based on the verified guidance:
- Keep Context Limit = 300K as the default for most day-to-day work.
- Use Max only when you’re doing broad, cross-solution work and you truly need long history/large retrieved context (expect higher latency/cost).
Workflow recommendations the docs explicitly imply:
- Prefer narrow scopes: scope the agent to only the relevant module/package/folder instead of the whole solution.
- Prefer targeted attachments: attach only the specific files/logs/errors needed for the current step.
- Avoid “carry everything forward” chats: for long-running tasks, periodically start a new session with a tight scope + only the artifacts needed (plan file + current failing error + list of remaining tasks).
3) Does Studio compact history / preserve checkpoints? How should a new session reconnect a saved plan?
No verified documentation was found that states:
- Studio automatically compacts/summarizes history, or
- Studio creates continuation checkpoints that a new chat can reliably resume from, or
- A guaranteed procedure for “reconnect saved plan file and continue unfinished steps” when the UI says “no active plan”.
So the only defensible, supported approach (based on what’s documented) is:
- Treat plans as files/artifacts you can re-feed to a new session via targeted attachment, and re-scope the agent to the specific area of code to continue.
- If Studio shows “no active plan” after reopening, attach the plan file content (or the relevant sections) to the new session and instruct the agent to continue from the remaining steps.
(No verified source that Studio has an automatic “resume from saved plan” mechanism across sessions in the way you described.)
4) After timeout/transport failure, how to know whether tool ops or file saves completed?
No verified public documentation was found describing:
- a deterministic “tool execution idempotency” mechanism,
- a “transaction log” for tool operations,
- or a UI that confirms “file save definitely completed” after a transport failure.
Practically (and safely) in ABP Studio workflows, the reliable approach is to verify outcomes from the workspace state before retrying:
- Check git diff / file timestamps / file contents to confirm changes landed.
- Re-run a build (or the specific deterministic step, like proxy generation/migration) to confirm the operation’s effect is present.
- If your process allows it, use smaller, more atomic agent steps (so after a failure you can quickly verify what changed).
(Again: no verified doc describing a built-in “exactly-once” guarantee for agent tool calls across network failures.)
5–6) Charges for failures, credit restoration, exporting IDs/records
These are billing/account-specific and not something I can answer definitively from the public docs/snippets retrieved. The supported route is:
- https://abp.io/contact
Ask them to:
- review whether the affected requests were charged,
- confirm retry/timeout charging behavior for your organization,
- and advise the official way to export/share session identifiers and usage records for investigation.
(light references)
- https://abp.io/docs/latest/studio/ai-agent-configuration
- https://abp.io/contact
This response was generated by AI. If it didn’t solve your problem, feel free to post a new answer — our support team will jump in to help you soon.
Markdown supported.Copy, paste, or drag & drop images and files (max 100 MB per file, 100 MB total per post) -
0
[AI-Bot] said:
- Licensing / billing / refunds / credit review (your items 5–6): please contact ABP via https://abp.io/contact (this is the supported/private route for billing review and account-specific investigations, including credit adjustments and sharing identifiers/logs).
- Technical behavior (your items 1–4): below is what’s verifiable from the current ABP Studio “AI Agent Configuration” documentation and generally supported Studio workflows.
1) Where are the “300K” vs “Max” settings in Studio 3.1.4, and what do they include?
- In ABP Studio, this is the AI Agent Configuration → Context Limit setting (the docs describe it as “controls the maximum context size used for agent runs”).
- 300K (default): Studio targets a maximum context size of ~300K for the agent run context.
- Max: “delegates to the model’s maximum context behavior.” So Studio will attempt to use whatever the selected model supports as its max context.
What the docs explicitly confirm:
- The limit applies to the agent run context (not just “chat messages”), and larger context increases latency and model cost.
- ABP recommends narrow AI scopes and targeted attachments to avoid large-context use when not needed.
What is not specified in the verified docs:
- A precise breakdown of whether the “context” number includes system instructions vs conversation history vs tool results vs reserved output vs internal reasoning tokens, or what Studio does if a chosen model supports less than 300K. (No verified public breakdown for this exact accounting.)
2) Recommended setting & workflow to avoid context errors without unnecessary credit usage
Based on the verified guidance:
- Keep Context Limit = 300K as the default for most day-to-day work.
- Use Max only when you’re doing broad, cross-solution work and you truly need long history/large retrieved context (expect higher latency/cost).
Workflow recommendations the docs explicitly imply:
- Prefer narrow scopes: scope the agent to only the relevant module/package/folder instead of the whole solution.
- Prefer targeted attachments: attach only the specific files/logs/errors needed for the current step.
- Avoid “carry everything forward” chats: for long-running tasks, periodically start a new session with a tight scope + only the artifacts needed (plan file + current failing error + list of remaining tasks).
3) Does Studio compact history / preserve checkpoints? How should a new session reconnect a saved plan?
No verified documentation was found that states:
- Studio automatically compacts/summarizes history, or
- Studio creates continuation checkpoints that a new chat can reliably resume from, or
- A guaranteed procedure for “reconnect saved plan file and continue unfinished steps” when the UI says “no active plan”.
So the only defensible, supported approach (based on what’s documented) is:
- Treat plans as files/artifacts you can re-feed to a new session via targeted attachment, and re-scope the agent to the specific area of code to continue.
- If Studio shows “no active plan” after reopening, attach the plan file content (or the relevant sections) to the new session and instruct the agent to continue from the remaining steps.
(No verified source that Studio has an automatic “resume from saved plan” mechanism across sessions in the way you described.)
4) After timeout/transport failure, how to know whether tool ops or file saves completed?
No verified public documentation was found describing:
- a deterministic “tool execution idempotency” mechanism,
- a “transaction log” for tool operations,
- or a UI that confirms “file save definitely completed” after a transport failure.
Practically (and safely) in ABP Studio workflows, the reliable approach is to verify outcomes from the workspace state before retrying:
- Check git diff / file timestamps / file contents to confirm changes landed.
- Re-run a build (or the specific deterministic step, like proxy generation/migration) to confirm the operation’s effect is present.
- If your process allows it, use smaller, more atomic agent steps (so after a failure you can quickly verify what changed).
(Again: no verified doc describing a built-in “exactly-once” guarantee for agent tool calls across network failures.)
5–6) Charges for failures, credit restoration, exporting IDs/records
These are billing/account-specific and not something I can answer definitively from the public docs/snippets retrieved. The supported route is:
- https://abp.io/contact
Ask them to:
- review whether the affected requests were charged,
- confirm retry/timeout charging behavior for your organization,
- and advise the official way to export/share session identifiers and usage records for investigation.
(light references)
- https://abp.io/docs/latest/studio/ai-agent-configuration
- https://abp.io/contact
This response was generated by AI. If it didn’t solve your problem, feel free to post a new answer — our support team will jump in to help you soon.
Thank you. The documentation summary is useful, but the version-specific questions remain unresolved.
Please assign this ticket to a Studio engineer who can confirm for Studio 3.1.4:
- How 300K and Max account for input, tools, reasoning and reserved output, and what happens when the selected model supports less than 300K.
- Whether Max changes reasoning effort automatically and whether context size and reasoning effort can be controlled separately.
- Whether automatic compaction/checkpoints exist, and the supported procedure for reconnecting an existing saved plan in a new session.
- How to identify completed tool operations after a timeout and export the relevant session/request identifiers.
We will verify saved artifacts before retrying. We will not rerun migrations or generators simply to determine whether they completed.
We will contact the billing team separately. Please keep this technical ticket open pending the engineering response.
Markdown supported.Copy, paste, or drag & drop images and files (max 100 MB per file, 100 MB total per post)
