hermes-agent/hermes_cli
Teknium 7b76366552
feat(prompt-cache): cross-session 1h prefix cache for Claude on Anthropic / OpenRouter / Nous Portal (#23828)
Cuts input cost for first-turn Claude requests by ~85-90% on subsequent
sessions within an hour. Tools array (~13k tokens for default toolset) +
stable system prefix (~5-8k tokens) get a 1h cache_control marker; the
volatile suffix (memory, USER profile, timestamp, session id) sits in a
separate non-cached block at the end so it doesn't poison the cross-session
prefix when it changes.

Provider gate: Claude on native Anthropic (incl. OAuth subscription),
OpenRouter, and Nous Portal (which proxies to OpenRouter). All other
providers keep today's system_and_3 layout unchanged.

Layout (4 cache_control breakpoints, Anthropic max):
  1. tools[-1]              -> 1h (cross-session)
  2. system content[0]      -> 1h (cross-session, stable prefix)
  3. messages[-2]           -> 5m (within-session rolling)
  4. messages[-1]           -> 5m (within-session rolling)

Within-session rolling shrinks from 3 messages to 2 to free the breakpoint
budget. On Claude with realistic tool loadouts the long-lived tier carries
the bulk of cross-session value anyway.

System prompt is now always assembled cache-friendly: stable identity /
guidance / skills / platform hints first, then session-stable context
files (AGENTS.md, .cursorrules), then per-call volatile content. Old
single-string callers see the same logical content (same join order),
just reordered so volatile lives at the end.

Config knobs (defaults shown):
  prompt_caching:
    cache_ttl: "5m"           # rolling-window TTL (unchanged)
    long_lived_prefix: true    # opt-out switch
    long_lived_ttl: "1h"       # cross-session prefix TTL

Live E2E (tests/agent/test_prompt_caching_live.py, gated on
OPENROUTER_API_KEY) on anthropic/claude-haiku-4.5 with default toolset:
  Call 1 (cold):              cache_write=13,415  cache_read=0
  Call 2 (NEW agent + msg):   cache_write=391     cache_read=13,025
  Cross-session reuse:        97.09%

Implementation:
* agent/prompt_caching.py: new apply_anthropic_cache_control_long_lived()
  + mark_tools_for_long_lived_cache(); existing apply_anthropic_cache_control()
  preserved verbatim for the fallback path.
* agent/anthropic_adapter.py: convert_tools_to_anthropic() now forwards
  cache_control onto each Anthropic-format tool dict.
* run_agent.py: _build_system_prompt_parts() returns the 3-tier dict;
  _build_system_prompt() joins them (backward compatible).
  _supports_long_lived_anthropic_cache() policy added next to the existing
  _anthropic_prompt_cache_policy() (which now also recognises Nous Portal
  Claude — pre-existing gap fixed in passing).
  _build_api_kwargs() resolves tools_for_api once and propagates the
  marker through all four build paths (anthropic_messages, bedrock,
  codex_responses, profile/legacy chat completions).
  Long-lived flag plumbed into the runtime snapshot/restore + model-switch
  + fallback-promotion paths.

Tests:
* tests/agent/test_prompt_caching.py: +8 tests (TestMarkToolsForLongLivedCache,
  TestApplyAnthropicCacheControlLongLived).
* tests/run_agent/test_anthropic_prompt_cache_policy.py: +9 tests
  (TestSupportsLongLivedAnthropicCache matrix across 8 endpoint classes
  + a fallback-target case).
* tests/agent/test_prompt_caching_live.py: new live E2E (skipif when
  OPENROUTER_API_KEY is unset; runs outside the hermetic suite).
* Targeted suites: 327/327 pass (caching/adapter/policy/builder).
* tests/agent/ + tests/run_agent/: 3992 pass, 17 skip, 1 pre-existing
  flake (test_async_httpx_del_neuter::test_same_key_replaces_stale_loop_entry,
  verified failing on pristine origin/main).
2026-05-11 11:14:56 -07:00
..
__init__.py chore: release v0.13.0 (2026.5.7) (#21406) 2026-05-07 09:22:48 -07:00
_parser.py fix: add dashboard to CLI help epilogue and Docker CI smoke test 2026-05-07 06:16:23 -07:00
_subprocess_compat.py feat(windows): close remaining POSIX-only landmines — TUI crash, kanban waitpid, AF_UNIX sandbox, /bin/bash, npm .cmd shims, cwd tracking, detach flags 2026-05-08 14:27:40 -07:00
auth.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
auth_commands.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
azure_detect.py chore: remove unused imports and dead locals (ruff F401, F841) (#17010) 2026-04-28 06:46:45 -07:00
backup.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
banner.py fix(banner): resolve update-check repo from running code, not profile-scoped path 2026-05-09 04:10:35 -07:00
browser_connect.py fix(browser): address Copilot review on /browser connect 2026-04-28 22:11:10 -07:00
callbacks.py fix: ESC cancels secret/sudo prompts, clearer skip messaging (#9902) 2026-04-14 16:11:37 -07:00
checkpoints.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
claw.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
cli_output.py refactor: remove dead code — 1,784 lines across 77 files (#9180) 2026-04-13 16:32:04 -07:00
clipboard.py feat: fix img pasting in new ink plus newline after tools 2026-04-11 13:14:32 -05:00
codex_models.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
colors.py feat: respect NO_COLOR env var and TERM=dumb (#4079) 2026-03-30 17:07:21 -07:00
commands.py chore: ruff auto-fixes — collapsible-else-if, if-stmt-min-max, dict.fromkeys (#23926) 2026-05-11 11:03:29 -07:00
completion.py fix(completion): use valid zsh _arguments exclusion-group syntax 2026-05-09 13:36:44 -07:00
config.py feat(prompt-cache): cross-session 1h prefix cache for Claude on Anthropic / OpenRouter / Nous Portal (#23828) 2026-05-11 11:14:56 -07:00
copilot_auth.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
cron.py feat(cron): add no_agent mode for script-only cron jobs (watchdog pattern) (#19709) 2026-05-04 12:31:01 -07:00
curator.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
curses_ui.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
debug.py fix(debug): redact log content at upload time in hermes debug share 2026-05-03 11:42:20 -07:00
default_soul.py fix: reset default SOUL.md to baseline identity text (#3159) 2026-03-26 01:34:27 -07:00
dingtalk_auth.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
doctor.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
dump.py refactor(env): use shared Hermes dotenv loader 2026-05-05 10:13:13 -07:00
env_loader.py feat(cross-platform): psutil for PID/process management + Windows footgun checker 2026-05-08 14:27:40 -07:00
fallback_cmd.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
gateway.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
gateway_windows.py fix(gateway): preserve Ctrl+C for Windows foreground runs 2026-05-09 14:34:18 -07:00
goals.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
hooks.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
kanban.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
kanban_db.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
kanban_diagnostics.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
kanban_specify.py feat(kanban): add specify — auxiliary LLM fleshes out triage tasks (#21435) 2026-05-07 13:04:41 -07:00
logs.py feat: component-separated logging with session context and filtering (#7991) 2026-04-11 17:23:36 -07:00
main.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
mcp_config.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
memory_setup.py codebase: add encoding='utf-8' to all bare open() calls (PLW1514) 2026-05-08 14:27:40 -07:00
model_catalog.py codebase: add encoding='utf-8' to all bare open() calls (PLW1514) 2026-05-08 14:27:40 -07:00
model_normalize.py fix(opencode-go): keep users on opencode-go instead of hijacking to native providers (#20802) 2026-05-06 09:08:33 -07:00
model_switch.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
models.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
nous_subscription.py feat(web): add SearXNG as a native search-only backend 2026-05-06 10:05:29 -07:00
oneshot.py fix: make session search initialize session db 2026-05-09 14:36:58 -07:00
pairing.py fix(pairing): enforce lockout on approve_code, not just generate_code (#10195) (#21325) 2026-05-07 07:18:21 -07:00
platforms.py feat: complete plugin platform parity — all 12 integration points 2026-04-29 21:56:51 -07:00
plugins.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
plugins_cmd.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
profile_distribution.py feat(profile): shareable profile distributions via git (#20831) 2026-05-08 10:04:32 -07:00
profiles.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
providers.py fix: prevent bare 'custom' slug in model.provider (#17478) 2026-04-30 04:32:11 -07:00
pt_input_extras.py fix(cli): make Ctrl+Enter insert newline on WSL/SSH/Windows Terminal (#22777) 2026-05-09 12:48:14 -07:00
pty_bridge.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
relaunch.py fix(windows): prefer npm.cmd over npm.ps1, skip .py argv0 in relaunch 2026-05-08 14:27:40 -07:00
runtime_provider.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
setup.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
skills_config.py refactor(config): migrate remaining 33 cfg_get call sites (#17311) 2026-04-29 04:03:03 -07:00
skills_hub.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
skin_engine.py fix(tui): honor skin highlight colors (#20895) 2026-05-06 14:01:56 -07:00
slack_cli.py fix(slack): enable writable app home DMs in manifest 2026-05-08 17:01:12 -07:00
status.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
stdio.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
timeouts.py refactor(timeouts): drop redundant ImportError in except clause 2026-04-26 20:48:20 -07:00
tips.py feat: Ctrl+Enter inserts newline on Windows Terminal 2026-05-08 14:27:40 -07:00
tools_config.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
uninstall.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
vercel_auth.py feat: add Vercel Sandbox backend 2026-04-29 07:22:33 -07:00
voice.py fix(tui): restore voice push-to-talk parity (#20897) 2026-05-06 15:49:59 -07:00
web_server.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00
webhook.py chore: ruff auto-fix PLR6201 — tuple → set in membership tests (#23937) 2026-05-11 11:13:25 -07:00