Atlas fails with Tool call not allowed while generating summary: <toolName> because the session processor rejects tool-input-start and tool-call stream events whenever the assistant message is a summary message, so a summarization pass cannot start editing files or running commands. The fix is to re-run the summarization without tools, since this is a guard and not a transient failure, and to perform the tool work in a normal, non-summary turn instead.
Why does Atlas say Tool call not allowed while generating summary
Atlas says Tool call not allowed while generating summary: <toolName> because the session processor rejects tool-input-start and tool-call stream events whenever the assistant message is a summary message. The 2026 guard fires the moment the model tries to act during a summarization pass.
A summary turn exists to compress what already happened, not to do new work. The Atlas session processor enforces that by watching the stream: if a tool-input-start or tool-call event arrives while the assistant message is marked as a summary message, the processor rejects it and names the offending tool in the error. The named <toolName> in the message tells you exactly which tool the model reached for, which is usually enough to see why: a model asked to summarize a session full of edits often decides it should keep editing. The guard stops that before a file is touched.
How to fix Tool call not allowed while generating summary in Atlas
Fix Tool call not allowed while generating summary in Atlas by re-running the summarization with 0 tools available. The Atlas session processor rejects tool-call events on a summary message by design, so this is a guard, not a transient failure, and retrying the identical request will fail identically.
Do not treat the error as flaky and retry it unchanged. The rejection is deterministic: as long as the assistant message is a summary message and the model emits a tool-call, the session processor will refuse it. Re-issue the summarization in a form that gives the model nothing to call. If the summary genuinely needs information the model does not have, gather that information first in a normal turn, then summarize afterward with the material already in context. Splitting the two phases is the durable fix, not a workaround.
How to remove tools from a custom summarizing agent in Atlas
If a custom Atlas agent is doing the summarizing, remove tools from its permission set so it has 0 tools to call. Every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, so an agent with nothing available never emits the tool-call event that triggers the error.
A custom summarizer that inherits a full tool set is an invitation for the model to reach for one. Deny the tools in the agent's permission rules so the summarization path has nothing to invoke. That aligns the agent's capabilities with its job: a summarizer reads and compresses, it does not edit files or run commands. The change is worth making even where the guard already protects you, because a rejected tool-call still burns a turn and produces an error the user has to interpret, while an agent with no tools simply writes the summary.
Why an Atlas turn cannot both summarize and act
An Atlas turn cannot both summarize and act, because the session processor rejects 2 stream event types, tool-input-start and tool-call, whenever the assistant message is a summary message. Check that the model is not being asked to do both in the same turn, since that request is unsatisfiable by construction.
Prompts that say something like summarize the session and then apply the remaining fixes will always fail on the second half. The summary message is a distinct message type in the Atlas session processor, and the guard is attached to the message type, not to the wording of the request. Separate the intents. Ask for the summary in one turn and the work in the next. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, which is the same principle applied at the agent level: thinking about the work and doing the work are different modes.
How to run the tool work Atlas blocked during the summary
Perform the tool work in a normal, non-summary turn instead. Atlas blocks the call in exactly 1 condition, while the assistant message is a summary message, so the same tool named in Tool call not allowed while generating summary: <toolName> runs without complaint in an ordinary turn.
Take the tool the error named and issue the request again as regular work. Nothing about the tool itself is restricted: the restriction is scoped to summary messages. Once you are in a normal turn, the usual Atlas protections apply, since every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, and Atlas computes a unified diff for every file edit and surfaces it for approval before writing. The result is what you wanted in the first place, arrived at through the turn type that permits it.
How to fix it
- 01Re-run the summarization without tools; this is a guard, not a transient failure.
- 02If a custom agent is being used to summarize, remove tools from its permission set.
- 03Check that the model is not being asked to both summarize and act in the same turn.
- 04Perform the tool work in a normal (non-summary) turn instead.
Frequently asked questions
- how to fix Tool call not allowed while generating summary in Atlas
- Re-run the summarization without tools. The Atlas session processor rejects tool-call events on a summary message by design, so this is a guard, not a transient failure, and an unchanged retry will fail the same way.
- why does Atlas block tools during summarization
- The Atlas session processor rejects tool-input-start and tool-call stream events whenever the assistant message is a summary message, so a summarization pass cannot start editing files or running commands.
- is Tool call not allowed while generating summary a transient Atlas error
- No. It is a guard, not a transient failure. Retrying the same summarization with tools available will produce the same error every time.
- how do I stop my custom Atlas agent from calling tools while summarizing
- Remove tools from the agent's permission set. With no tools available, the summarization path cannot emit the tool-call event that triggers the error.
- can an Atlas turn summarize and edit files at the same time
- No. Check that the model is not being asked to both summarize and act in the same turn. Perform the tool work in a normal, non-summary turn instead.
- what does the toolName in the Atlas summary error mean
- Tool call not allowed while generating summary: <toolName> names the exact tool the model tried to call during the summary message, which usually explains why the guard fired.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas with Kimi K2.7 Code HighSpeed: The 2x Throughput Tier in 2026
Kimi K2.7 Code HighSpeed runs Atlas at $1.90 per Mtok input and $8.00 per Mtok output, exactly 2x K2.7 Code, on the same 256K tokens (262,144) context window.
Atlas with DeepSeek R1 (0528): The Open Reasoning Trace, 2026
Run Atlas on DeepSeek R1 (0528) in 2026. DeepInfra hosts the MIT-licensed open reasoning model at $0.50 per Mtok in, $2.15 per Mtok out, 160K tokens context.
Audit a Repo with Parallel Subagents in Atlas (2026 Workflow)
How to audit a repo with parallel subagents in Atlas in 2026: the task tool launches explore subagents in their own sessions, so only conclusions return to your context.
Atlas vs JetBrains Junie in 2026: Terminal Agent vs IDE Debugger-Driving Agent
Atlas vs JetBrains Junie in 2026: Junie drives the IDE debugger and reports 61.6% resolved on SWE-Rebench, bundled from $10/mo. Atlas is a free, open source terminal agent with permission gates.
Atlas vs Bolt.new in 2026: Terminal Agent or In-Browser WebContainer Builder
Atlas is a free, open source terminal-native AI coding agent. Bolt.new runs npm install and your dev server in-browser via WebContainers. Compared for 2026.
Atlas for Java in 2026
Adopt Atlas, the terminal-native AI coding agent, for Java development in 2026. Enhance your workflow with intelligent code search, refactoring, and robust safety features for Maven and Gradle projects.
Atlas with Mistral Medium 3 (2505) in 2026: Symmetric Limits, $0.40 In
Mistral Medium 3 (2505) runs Atlas with symmetric 131,072 token context and output at $0.40 / 1M input tokens and $2.00 / 1M output tokens. Setup, limits, successors.
Atlas for Flutter in 2026
Discover Atlas for Flutter in 2026. This terminal-native AI coding agent helps Flutter developers build apps faster and safer, integrating with widgets, state, and the Dart toolchain.