Nanobot isolates multi-client API session state and adds automated self-update flows

Nanobot developers submitted PR #5838 on September 20, 2026, resolving a major concurrency flaw in the agent's REST API layer. When multiple external clients or automated services dispatched requests to the Nanobot API server concurrently, incoming queries were frequently multiplexed into a shared default chat context rather than isolating each session_id. This caused interleaved message histories, context window contamination across independent users, and catastrophic prompt leakage in multi-tenant environments. PR #5838 refactors request routing to instantiate and maintain discrete session contexts bound strictly to the calling session_id. In parallel, PR #5817 introduces automated self-update workflows for both stable PyPI releases and source checkouts, giving headless server installations a unified upgrade mechanism. Concurrently, PR #5769 improves provider resilience by adding automatic failover logic for NVIDIA NIM timeout errors, preventing gateway hangs when enterprise inference clusters throttle requests. Meanwhile, issue #5833 highlights that Nanobot's SSE streaming consumer drops reasoning_text events that are properly handled by the direct SDK consumer, truncating internal model deliberation during live chat.

The facts

- PR #5838 routes each API session_id to its own independent chat context, eliminating cross-client context leakage. - Prevents multi-user deployments from interleaving dialogue histories and exposing sensitive prompt context across sessions. - PR #5817 adds native self-update commands for both packaged PyPI releases and git-based source installations. - PR #5769 introduces automated provider failover handling for NVIDIA NIM-style timeout exceptions during model calls. - Issue #5833 reports that the SSE Responses consumer silently discards response.reasoning_text events during streamed turns.

Why it matters

API server session isolation is a non-negotiable security and reliability requirement for any agent framework intended for multi-tenant deployment. If concurrent HTTP requests bleed context into a shared conversation buffer, confidential user inputs are leaked and model reasoning is derailed by irrelevant external turns. Binding session lifecycles strictly to client identifiers restores session confidentiality. Concurrently, automated self-update tooling and robust inference timeout failovers lower administrative overhead for production installations.

Current

Inspected on 2026-09-21. The Nanobot stable-channel baseline is v0.3.5 published 2026-09-15T19:05:37Z. The primary source was open (not merged) when captured. Mainline merges, open proposals and packaged releases are distinct availability states; the release baseline does not establish that a proposal has shipped.

Evidence

Primary evidence: HKUDS/nanobot PR #5838 (https://github.com/HKUDS/nanobot/pull/5838). Supporting context: HKUDS/nanobot PR #5817 (https://github.com/HKUDS/nanobot/pull/5817); HKUDS/nanobot PR #5769 (https://github.com/HKUDS/nanobot/pull/5769); HKUDS/nanobot issue #5833 (https://github.com/HKUDS/nanobot/issues/5833). The source bodies and linked context were inspected. Test results quoted from contributors remain attributed reports, not independently reproduced experiments.

Operator take

Never assume statelessness in multi-client API bridges without explicit session routing tests. Agent frameworks must treat concurrent session isolation as a primary security boundary. Organizations hosting Nanobot as an internal microservice should upgrade to versions incorporating PR #5838 to guarantee tenant boundary enforcement and configure provider failover lists to survive upstream model timeouts.

Caveat

PR #5838 is currently open for code review. Operators running multi-user API endpoints should verify that clients supply unique, cryptographically random session_id headers to prevent intentional or accidental session collision.