Previously, when a session expired (idle/daily reset), the memory flush
ran synchronously inside get_or_create_session — blocking the user's
message for 10-60s while an LLM call saved memories.
Now a background watcher task (_session_expiry_watcher) runs every 5 min,
detects expired sessions, and flushes memories proactively in a thread
pool. By the time the user sends their next message, memories are
already saved and the response is immediate.
Changes:
- Add _is_session_expired(entry) to SessionStore — works from entry
alone without needing a SessionSource
- Add _pre_flushed_sessions set to track already-flushed sessions
- Remove sync _on_auto_reset callback from get_or_create_session
- Refactor flush into _flush_memories_for_session (sync worker) +
_async_flush_memories (thread pool wrapper)
- Add _session_expiry_watcher background task, started in start()
- Simplify /reset command to use shared fire-and-forget flush
- Add 10 tests for expiry detection, callback removal, tracking
Reduces token usage and latency for most tasks by defaulting to
medium reasoning effort instead of xhigh. Users can still override
via config or CLI flag. Updates code, tests, example config, and docs.
_make_cli() did not clear HERMES_MAX_ITERATIONS env var, so tests
failed in CI where the var was set externally. Also, default max_turns
changed from 60 to 90 in 0a82396 but tests were not updated.
- Clear HERMES_MAX_ITERATIONS in _make_cli() for proper isolation
- Add env_overrides parameter for tests that need specific env values
- Update hardcoded 60 assertions to 90 to match new default
- Simplify test_env_var_max_turns using env_overrides
Add a 'platforms' field to SKILL.md frontmatter that restricts skills
to specific operating systems. Skills with platforms: [macos] only
appear in the system prompt, skills_list(), and slash commands on macOS.
Skills without the field load everywhere (backward compatible).
Implementation:
- skill_matches_platform() in tools/skills_tool.py — core filter
- Wired into all 3 discovery paths: prompt_builder.py, skills_tool.py,
skill_commands.py
- 28 new tests across 3 test files
New bundled Apple/macOS skills (all platforms: [macos]):
- imessage — Send/receive iMessages via imsg CLI
- apple-reminders — Manage Reminders via remindctl CLI
- apple-notes — Manage Notes via memo CLI
- findmy — Track devices/AirTags via AppleScript + screen capture
Docs updated: CONTRIBUTING.md, AGENTS.md, creating-skills.md,
skills.md (user guide)
Authored by areu01or00. Adds timezone support via hermes_time.now() helper
with IANA timezone resolution (HERMES_TIMEZONE env → config.yaml → server-local).
Updates system prompt timestamp, cron scheduling, and execute_code sandbox TZ
injection. Includes config migration (v4→v5) and comprehensive test coverage.
Adds 4 new direct API-key providers (zai, kimi-coding, minimax, minimax-cn)
to the inference provider system. All use standard OpenAI-compatible
chat/completions endpoints with Bearer token auth.
Core changes:
- auth.py: Extended ProviderConfig with api_key_env_vars and base_url_env_var
fields. Added providers to PROVIDER_REGISTRY. Added provider aliases
(glm, z-ai, zhipu, kimi, moonshot). Added auto-detection of API-key
providers in resolve_provider(). Added resolve_api_key_provider_credentials()
and get_api_key_provider_status() helpers.
- runtime_provider.py: Added generic API-key provider branch in
resolve_runtime_provider() — any provider with auth_type='api_key'
is automatically handled.
- main.py: Added providers to hermes model menu with generic
_model_flow_api_key_provider() flow. Updated _has_any_provider_configured()
to check all provider env vars. Updated argparse --provider choices.
- setup.py: Added providers to setup wizard with API key prompts and
curated model lists.
- config.py: Added env vars (GLM_API_KEY, KIMI_API_KEY, MINIMAX_API_KEY,
etc.) to OPTIONAL_ENV_VARS.
- status.py: Added API key display and provider status section.
- doctor.py: Added connectivity checks for each provider endpoint.
- cli.py: Updated provider docstrings.
Docs: Updated README.md, .env.example, cli-config.yaml.example,
cli-commands.md, environment-variables.md, configuration.md.
Tests: 50 new tests covering registry, aliases, resolution, auto-detection,
credential resolution, and runtime provider dispatch.
Inspired by PR #33 (numman-ali) which proposed a provider registry approach.
Credit to tars90percent (PR #473) and manuelschipper (PR #420) for related
provider improvements merged earlier in this changeset.
API key selection is now base_url-aware: when the resolved base_url
targets OpenRouter, OPENROUTER_API_KEY takes priority (preserving the
#289 fix). When hitting any other endpoint (Z.ai, vLLM, custom, etc.),
OPENAI_API_KEY takes priority so the OpenRouter key doesn't leak.
Applied in both the runtime provider resolver (the real code path) and
the CLI initial default (for consistency).
Fixes#560.
_make_cli() now patches CLI_CONFIG with clean defaults so
test_cli_init tests don't depend on the developer's local config.yaml.
test_empty_dir_returns_empty now mocks Path.home() so it doesn't pick
up a global SOUL.md.
Credit to teyrebaz33 for identifying and fixing these in PR #557.
Fixes#555.
Two bugs in sync_skills():
1. Failed copytree poisons manifest: when shutil.copytree fails (disk
full, permission error), the skill is still recorded in the manifest.
On the next sync, the skill appears as "in manifest but not on disk"
which is interpreted as "user deliberately deleted it" — the skill
is never retried. Fix: only write to manifest on successful copy.
2. Failed update destroys user copy: rmtree deletes the existing skill
directory before copytree runs. If copytree then fails, the user's
skill is gone with no way to recover. Fix: move to .bak before
copying, restore from backup if copytree fails.
Both bugs are proven by new regression tests that fail on the old code
and pass on the fix.
Upgrade skills_sync manifest to v2 format (name:origin_hash). The origin
hash records the MD5 of the bundled skill at the time it was last synced.
On update, the user's copy is compared against the origin hash:
- User copy == origin hash → unmodified → safe to update from bundled
- User copy != origin hash → user customized → skip (preserve changes)
v1 manifests (plain names) are auto-migrated: the user's current hash
becomes the baseline, so future syncs can detect modifications.
Output now shows user-modified skills:
~ whisper (user-modified, skipping)
27 tests covering all scenarios including v1→v2 migration, user
modification detection, update after migration, and origin hash tracking.
2009 tests pass.
- Restored 21 skills removed in commits 757d012 and 740dd92:
accelerate, audiocraft, code-review, faiss, flash-attention, gguf,
grpo-rl-training, guidance, llava, nemo-curator, obliteratus, peft,
pytorch-fsdp, pytorch-lightning, simpo, slime, stable-diffusion,
tensorrt-llm, torchtitan, trl-fine-tuning, whisper
- Rewrote sync_skills() with proper update semantics:
* New skills (not in manifest): copied to user dir
* Existing skills (in manifest + on disk): updated via hash comparison
* User-deleted skills (in manifest, not on disk): respected, not re-added
* Stale manifest entries (removed from bundled): cleaned from manifest
- Added sync_skills() to CLI startup (cmd_chat) and gateway startup
(start_gateway) — previously only ran during 'hermes update'
- Updated cmd_update output to show new/updated/cleaned counts
- Rewrote tests: 20 tests covering manifest CRUD, dir hashing, fresh
install, user deletion respect, update detection, stale cleanup, and
name collision handling
75 bundled skills total. 2002 tests pass.
Issues found and fixed during deep code path review:
1. CRITICAL: Prefix matching returned wrong prices for dated model names
- 'gpt-4o-mini-2024-07-18' matched gpt-4o ($2.50) instead of gpt-4o-mini ($0.15)
- Same for o3-mini→o3 (9x), gpt-4.1-mini→gpt-4.1 (5x), gpt-4.1-nano→gpt-4.1 (20x)
- Fix: use longest-match-wins strategy instead of first-match
- Removed dangerous key.startswith(bare) reverse matching
2. CRITICAL: Top Tools section was empty for CLI sessions
- run_agent.py doesn't set tool_name on tool response messages (pre-existing)
- Insights now also extracts tool names from tool_calls JSON on assistant
messages, which IS populated for all sessions
- Uses max() merge strategy to avoid double-counting between sources
3. SELECT * replaced with explicit column list
- Skips system_prompt and model_config blobs (can be thousands of chars)
- Reduces memory and I/O for large session counts
4. Sets in overview dict converted to sorted lists
- models_with_pricing / models_without_pricing were Python sets
- Sets aren't JSON-serializable — would crash json.dumps()
5. Negative duration guard
- end > start check prevents negative durations from clock drift
6. Model breakdown sort fallback
- When all tokens are 0, now sorts by session count instead of arbitrary order
7. Removed unused timedelta import
Added 6 new tests: dated model pricing (4), tool_calls JSON extraction,
JSON serialization safety. Total: 69 tests.
Custom OAI endpoints, self-hosted models, and local inference should NOT
show fabricated cost estimates. Changed default pricing from $3/$12 per
million tokens to $0/$0 for unrecognized models.
- Added _has_known_pricing() to distinguish commercial vs custom models
- Models with known pricing show $ amounts; unknown models show 'N/A'
- Overview shows asterisk + note when some models lack pricing data
- Gateway format adds '(excludes custom/self-hosted models)' note
- Added 7 new tests for custom model cost handling
Inspired by Claude Code's /insights, adapted for Hermes Agent's multi-platform
architecture. Analyzes session history from state.db to produce comprehensive
usage insights.
Features:
- Overview stats: sessions, messages, tokens, estimated cost, active time
- Model breakdown: per-model sessions, tokens, and cost estimation
- Platform breakdown: CLI vs Telegram vs Discord etc. (unique to Hermes)
- Tool usage ranking: most-used tools with percentages
- Activity patterns: day-of-week chart, peak hours, streaks
- Notable sessions: longest, most messages, most tokens, most tool calls
- Cost estimation: real pricing data for 25+ models (OpenAI, Anthropic,
DeepSeek, Google, Meta) with fuzzy model name matching
- Configurable time window: --days flag (default 30)
- Source filtering: --source flag to filter by platform
Three entry points:
- /insights slash command in CLI (supports --days and --source flags)
- /insights slash command in gateway (compact markdown format)
- hermes insights CLI subcommand (standalone)
Includes 56 tests covering pricing helpers, format helpers, empty DB,
populated DB with multi-platform data, filtering, formatting, and edge cases.
Authored by Farukest. Fixes#432. Extracts _kill_port_process() helper
that uses netstat+taskkill on Windows and fuser on Linux. Previously,
fuser calls were inline with bare except-pass, so on Windows orphaned
bridge processes were never cleaned up — causing 'address already in use'
errors on reconnect. Includes 5 tests covering both platforms, port
matching edge cases, and exception suppression.
Authored by Farukest. Fixes#435. The retry summary in
_handle_max_iterations() hardcoded max_tokens instead of using
_max_tokens_param(), which returns max_completion_tokens for direct
OpenAI API (required by gpt-4o, o-series). The first attempt already
used _max_tokens_param correctly — only the retry path was wrong.
Includes 4 tests for _max_tokens_param provider detection.
Verifies explicit allowlist keys, catch-all _API_KEY/_TOKEN patterns,
case insensitivity, TERMINAL_SSH prefix, and config.yaml routing for
non-secret keys. Covers the fix from PR #469.
The mock handler checked for function_name == 'search' but the RPC
sends 'search_files'. Any test exercising search_files through the
mock would get 'Unknown tool' instead of the canned response.
The _TOOL_STUBS dict in code_execution_tool.py was out of sync with the
actual tool schemas, causing TypeErrors when the LLM used parameters it
sees in its system prompt but the sandbox stubs didn't accept:
search_files:
- Added missing params: context, offset, output_mode
- Fixed target default: 'grep' → 'content' (old value was obsolete)
patch:
- Added missing params: mode, patch (V4A multi-file patch support)
Also added 4 drift-detection tests (TestStubSchemaDrift) that will
catch future divergence between stubs and real schemas:
- test_stubs_cover_all_schema_params: every schema param in stub
- test_stubs_pass_all_params_to_rpc: every stub param sent over RPC
- test_search_files_target_uses_current_values: no obsolete values
- test_generated_module_accepts_all_params: generated code compiles
All 28 tests pass.
Authored by rovle. Adds Daytona as the sixth terminal execution backend
with cloud sandboxes, persistent workspaces, and full CLI/gateway integration.
Includes 24 unit tests and 8 integration tests.
The execute_code sandbox generates a hermes_tools.py stub module for LLM
scripts. Three common failure modes keep tripping up scripts:
1. json.loads(strict=True) rejects control chars in terminal() output
(e.g., GitHub issue bodies with literal tabs/newlines)
2. Shell backtick/quote interpretation when interpolating dynamic content
into terminal() commands (markdown with backticks gets eaten by bash)
3. No retry logic for transient network failures (API timeouts, rate limits)
Adds three convenience helpers to the generated hermes_tools module:
- json_parse(text) — json.loads with strict=False for tolerant parsing
- shell_quote(s) — shlex.quote() for safe shell interpolation
- retry(fn, max_attempts=3, delay=2) — exponential backoff wrapper
Also updates the EXECUTE_CODE_SCHEMA description to document these helpers
so LLMs know they're available without importing anything extra.
Includes 7 new tests (unit + integration) covering all three helpers.
The original implementation only supported xclip (X11), which silently
fails on WSL2 (can't access Windows clipboard for images), Wayland
desktops (xclip is X11-only), and VSCode terminal on WSL2.
Clipboard backend changes (hermes_cli/clipboard.py):
- WSL2: detect via /proc/version, use powershell.exe with .NET
System.Windows.Forms.Clipboard to extract images as base64 PNG
- Wayland: use wl-paste with MIME type detection, auto-convert BMP
to PNG for WSLg environments (via Pillow or ImageMagick)
- Dispatch order: WSL → Wayland → X11 (xclip), with fallthrough
- New has_clipboard_image() for lightweight clipboard checks
- Cache WSL detection result per-process
CLI changes (cli.py):
- /paste command: explicit clipboard image check for terminals where
BracketedPaste doesn't fire (image-only clipboard in VSCode/WinTerm)
- Ctrl+V keybinding: fallback for Linux terminals where Ctrl+V sends
raw byte instead of triggering bracketed paste
Tests: 80 tests (up from 37) covering WSL, Wayland, X11 dispatch,
BMP conversion, has_clipboard_image, and /paste command.
Copy an image to clipboard (screenshot, browser, etc.) and paste into
the Hermes CLI. The image is saved to ~/.hermes/images/, shown as a
badge above the input ([📎 Image #1]), and sent to the model as a
base64-encoded OpenAI vision multimodal content block.
Implementation:
- hermes_cli/clipboard.py: clean module with platform-specific extraction
- macOS: pngpaste (if installed) → osascript fallback (always available)
- Linux: xclip (apt install xclip)
- cli.py: BracketedPaste key handler checks clipboard on every paste,
image bar widget shows attached images, chat() converts to multimodal
content format, Ctrl+C clears attachments
Inspired by @m0at's fork (https://github.com/m0at/hermes-agent) which
implemented image paste support for local vision models. Reimplemented
cleanly as a separate module with tests.
On top of PR #460: self-hosted Firecrawl instances don't require an API
key (USE_DB_AUTHENTICATION=false), so don't force users to set a dummy
FIRECRAWL_API_KEY when FIRECRAWL_API_URL is set. Also adds a proper
self-hosting section to the configuration docs explaining what you get,
what you lose, and how to set it up (Docker stack, tradeoffs vs cloud).
Added 2 more tests (URL-only without key, neither-set raises).
Replaces the unsafe 128K fallback for unknown models with a descending
probe strategy (2M → 1M → 512K → 200K → 128K → 64K → 32K). When a
context-length error occurs, the agent steps down tiers and retries.
The discovered limit is cached per model+provider combo in
~/.hermes/context_length_cache.yaml so subsequent sessions skip probing.
Also parses API error messages to extract the actual context limit
(e.g. 'maximum context length is 32768 tokens') for instant resolution.
The CLI banner now displays the context window size next to the model
name (e.g. 'claude-opus-4 · 200K context · Nous Research').
Changes:
- agent/model_metadata.py: CONTEXT_PROBE_TIERS, persistent cache
(save/load/get), parse_context_limit_from_error(), get_next_probe_tier()
- agent/context_compressor.py: accepts base_url, passes to metadata
- run_agent.py: step-down logic in context error handler, caches on success
- cli.py + hermes_cli/banner.py: context length in welcome banner
- tests: 22 new tests for probing, parsing, and caching
Addresses #132. PR #319's approach (8K default) rejected — too conservative.
Adds optional FIRECRAWL_API_URL environment variable to support
self-hosted Firecrawl deployments alongside the cloud service.
- Add FIRECRAWL_API_URL to optional env vars in hermes_cli/config.py
- Update _get_firecrawl_client() in tools/web_tools.py to accept custom API URL
- Add tests for client initialization with/without URL
- Document new env var in installation and config guides
The Daytona SDK's process.exec(timeout=N) parameter is not enforced —
the server-side timeout never fires and the SDK has no client-side
fallback, causing commands to hang indefinitely.
Fix: wrap commands with timeout N sh -c '...' (coreutils) which
reliably kills the process and returns exit code 124. Added
shlex.quote for proper shell escaping and a secondary deadline (timeout + 10s) that force-stops the sandbox if the shell timeout somehow fails.
Signed-off-by: rovle <lovre.pesut@gmail.com>
state
- Replace logger.warning with warnings.warn for the disk cap so users
actually see it (logger was suppressed by CLI's log level config)
- Use SandboxState enum instead of string literals in
_ensure_sandbox_ready
Signed-off-by: rovle <lovre.pesut@gmail.com>
Authored by 0xbyt4. Wraps commands with unique fence markers to isolate real output
from shell init/exit noise (oh-my-zsh, macOS session restore, etc.). Falls back to
expanded pattern-based cleaning. Also fixes BSD find fallback and test module shadowing.
The retry summary in _handle_max_iterations hardcodes max_tokens instead
of calling _max_tokens_param(). For direct OpenAI API users (gpt-4o,
o-series), the correct parameter name is max_completion_tokens. The first
attempt at line 2697 already uses _max_tokens_param correctly but the
retry path at line 2743 was missed.
fuser command does not exist on Windows, causing orphaned bridge processes
to never be cleaned up. On crash recovery, the port stays occupied and the
next connect() fails with address-already-in-use.
Add _kill_port_process() helper that uses netstat+taskkill on Windows and
fuser on Linux/macOS. Replace both call sites in connect() and disconnect().
