refactor: unify transport dispatch + collapse normalize shims by kshitijk4poor · Pull Request #13862 · NousResearch/hermes-agent

kshitijk4poor · 2026-04-22T05:32:09Z

Summary

PR 7 of the provider transport refactor (#13473). Final consolidation PR — combines the planned PR 7 (unify dispatch) and PR 8 (runtime simplification). Closes cycle 1 of the refactor.

Changes

1. Consolidate 4 transport helpers → 1 generic _get_transport(api_mode=None)

Replace _get_anthropic_transport(), _get_codex_transport(), _get_chat_completions_transport(), _get_bedrock_transport() with one method using a shared dict cache. 22 call sites updated.

2. Collapse 65-line main normalize block → 7 lines

One _get_transport().normalize_response() call + one shared _nr_to_assistant_message() shim. The shim extracts provider_data fields (codex_reasoning_items, reasoning_content, reasoning_details, call_id, response_item_id) into the SimpleNamespace shape downstream expects.

3. Wire ALL remaining response.choices[0] accesses through transports

Zero response.choices[0] remaining in non-streaming run_agent.py code. Every response access goes through transport.normalize_response(). Sites wired:

Main normalize (L10701)
Truncation handler (L9483)
flush_memories chat/bedrock tool extraction (L7304)
Iteration summary (L8378) + retry (L8419)
Bedrock finish_reason (L9451)
Chat completions finish_reason + truncation check (L9455)

4. Remove scaffolding from PRs 1-6

normalize_anthropic_response_v2 deleted from anthropic_adapter.py (39 lines) — transport now calls v1 directly
test_anthropic_normalize_v2.py deleted (238 lines) — tested a wrapper that no longer exists
10 dead imports removed from run_agent.py: 8 codex adapter functions + OPENROUTER_BASE_URL + DEVELOPER_ROLE_MODELS
Dead comment referencing removed import updated

5. Transport lifecycle

Eagerly warm transport cache at __init__ (surfaces import errors early)
Invalidate transport cache at ALL 4 self.api_mode = mutation sites: switch_model(), fallback activation, fallback restore, AND transport recovery

Impact

run_agent.py:           11,988 → 11,944 (-44 lines)
anthropic_adapter.py:    1,601 → 1,562 (-39 lines)
test_normalize_v2.py:      238 → 0      (-238 lines)
test_multimodal:             3 → 4      (+1 line, import path fix)
anthropic transport:       129 → 150    (+21 lines, inlined v2 logic)

Net:  +145 / -444 = -299 lines

Test plan

2,576 run_agent + agent tests pass (3 pre-existing flaky failures)
Zero response.choices[0] in non-streaming code
Transport cache invalidation at all 4 api_mode mutation sites

Consolidate 4 per-transport lazy singleton helpers (_get_anthropic_transport, _get_codex_transport, _get_chat_completions_transport, _get_bedrock_transport) into one generic _get_transport(api_mode) with a shared dict cache. Collapse the 65-line main normalize block (3 api_mode branches, each with its own SimpleNamespace shim) into 7 lines: one _get_transport() call + one _nr_to_assistant_message() shared shim. The shim extracts provider_data fields (codex_reasoning_items, reasoning_details, call_id, response_item_id) into the SimpleNamespace shape downstream code expects. Wire chat_completions and bedrock_converse normalize through their transports for the first time — these were previously falling into the raw response.choices[0].message else branch. Remove 8 dead codex adapter imports that have zero callers after PRs 1-6. Transport lifecycle improvements: - Eagerly warm transport cache at __init__ (surfaces import errors early) - Invalidate transport cache on api_mode change (switch_model, fallback activation, fallback restore, transport recovery) — prevents stale transport after mid-session provider switch run_agent.py: -32 net lines (11,988 -> 11,956). PR 7 of the provider transport refactor.

The transport refactor (PRs #13862 ff.) added agent/transports/ as a sub-package but the setuptools packages.find include list only had "agent" (top-level files), not "agent.*" (sub-packages). pip install / Nix builds therefore ship run_agent.py (which now imports from agent.transports on every API call) but omit the transports directory entirely, causing: ModuleNotFoundError: No module named 'agent.transports' on every LLM call for packaged installs. Adds "agent.*" to match the existing pattern used by tools, gateway, tui_gateway, and plugins.

…odex_responses Follow-up for #13862 — the post-init api_mode upgrade at __init__ (direct OpenAI / gpt-5-requires-responses path) runs AFTER the eager transport warm. Clear the cache so the stale chat_completions entry is evicted. Cosmetic: correctness was already fine since _get_transport() keys by current api_mode, but this avoids leaving unused cache state behind.

The transport refactor (PRs NousResearch#13862 ff.) added agent/transports/ as a sub-package but the setuptools packages.find include list only had "agent" (top-level files), not "agent.*" (sub-packages). pip install / Nix builds therefore ship run_agent.py (which now imports from agent.transports on every API call) but omit the transports directory entirely, causing: ModuleNotFoundError: No module named 'agent.transports' on every LLM call for packaged installs. Adds "agent.*" to match the existing pattern used by tools, gateway, tui_gateway, and plugins.

…odex_responses Follow-up for NousResearch#13862 — the post-init api_mode upgrade at __init__ (direct OpenAI / gpt-5-requires-responses path) runs AFTER the eager transport warm. Clear the cache so the stale chat_completions entry is evicted. Cosmetic: correctness was already fine since _get_transport() keys by current api_mode, but this avoids leaving unused cache state behind.