Files
agent-estate-wiki/systems/tool-search.md
T
Hermes 341dda95dd stocktake 2026-11-11: 36 stale pages re-verified (0 stale); fallback chain rewritten 4/6 slots; CT333 'clawtest' naming claim corrected; nous 404 credits signature replaces refresh-token 401; estate shrank (VM 310 + CT601 gone)
- freshness: 36 stale / 93 fresh / 3 no-date -> 0 stale / 129 fresh / 3 no-date of 132 (same-day report regenerated after the corrections, per pattern 13)
- skills: 299 skills / 67 categories (count unchanged; generator arithmetic re-verified by hand against the live tree: 300 SKILL.md files - 1 nested duplicate = 299)
- runbooks/test-openrouter: live fallback chain rewritten (primary back to smart-route/custom:omni-101; mimo-v2.5, stepfun, qwen3.5/ollama-cloud out; guacano + kilo-reflection-beam-free in); ollama.com dropped from the URL test list; guacano/litellm noted as internal + key-gated
- comparisons/hermes-vs-nanobot: the 2026-10-20 'CT333 is zeroclaw in Proxmox naming' claim was itself wrong - pct list and hostname both return clawtest
- incidents/2026-10-22-nous: refresh-token failure gone from logs; new signature 404 insufficient_credits_for_paid_model from auxiliary.pulse_memory (ninth slot, back on nous, paid model) - open item revised
- incidents/2026-08-29-gmail: email crons now ok but root cause unfixed 74 days (himalaya message read still in fetch_newsletters.py; goimapnotify back up PID 441; monitor still paused)
- 30 further pages re-verified live and de-flagged with dated probe notes (ct207 ingress, CT205 guanaco v0.8.15, CT270 nextcloud 33.0.2.2, headless stack, hermes-ui both hosts, systemd units + cgroup limits, terminal/tool_search/mnemosyne/custom_providers config blocks, ops profile still deleted, hosts/pve guest lists)
- log.md addendum + index.md date bump; no page deleted, no secrets written
2026-11-11 03:52:48 +08:00

104 lines
3.7 KiB
Markdown

---
title: Hermes Tool Search
type: system
status: active
created: 2026-07-25
updated: 2026-11-11
review_after: 2026-11-20
verified_on: 2026-11-11
version_applies_to: Hermes Agent on CT460
confidence: high
tags: [hermes, automation, system]
sources: [live CT460 configuration and isolated CUA validation]
---
# Hermes Tool Search
## Why it is enabled
Hermes previously sent every external tool schema to the model on every turn.
That used a large part of the context window before the conversation began and
made compaction happen much sooner.
Native Hermes Tool Search is now always on. Core tools remain immediately
available. External tools are discovered only when they are relevant, then
described and called through the Tool Search bridge.
## Active configuration
File: `/home/hermes/.hermes/config.yaml`
```yaml
tools:
tool_search:
enabled: "on"
threshold_pct: 20
search_default_limit: 8
max_search_limit: 50
listing_max_tokens: 8
```
> **Live config re-read 2026-10-20** — the values above are the current ones in
> `/home/hermes/.hermes/config.yaml`. They have drifted from the original 2026-07-25 rollout:
> `threshold_pct` 10 → **20**, `search_default_limit` 5 → **8**, `max_search_limit` 20 → **50**,
> and `listing_max_tokens: 8` has been added. `enabled: "on"` is unchanged.
> The token-budget figures in "Verified result" below were measured under the *original* settings
> and have not been re-measured since the tuning.
`enabled: "on"` is the important setting: it makes Tool Search active on every
request instead of waiting for the tool schemas to cross an automatic token
threshold.
> **Config re-read 2026-11-11 — byte-for-byte unchanged.** `tools.tool_search` in
> `~/.hermes/config.yaml` still reads `enabled: "on"`, `threshold_pct: 20`,
> `search_default_limit: 8`, `max_search_limit: 50`, `listing_max_tokens: 8`. `tools:` carries no
> other key. Tool Search was exercised throughout this stocktake session.
## Verified result
- 80 external tools were deferred in the production gateway.
- Approximately 24,151 schema tokens were removed from the always-present
prompt.
- 36 core or directly visible tools remained available.
- Estimated fixed baseline fell from about 52,955 to 29,182 tokens, a reduction
of roughly 45%.
- In an isolated Computer Use test, 53 tools were discovered and the
`health_report` search, description, and call all succeeded.
- Native Tool Search tests passed: 39 tests.
Tool Search can add a small discovery call when an external tool is first
needed. It should not reduce answer quality: the model retains direct access to
core tools and can search for specialized tools by name or purpose.
## New tools and new skills
New external tools are automatically included in Tool Search after their tool
provider is loaded. They do not need to be manually assigned to a category.
Skills use a separate discovery system:
- `~/.hermes/skills/devops/my-skill/SKILL.md` is categorized as `devops`.
- A skill directly under `~/.hermes/skills/` has no category.
- The skill `name` and `description` come from YAML frontmatter in `SKILL.md`.
- Directory additions are noticed automatically. An in-place edit may take up
to 30 seconds to appear because of the skill cache.
Tool Search therefore does not decide skill categories. The first directory
below `~/.hermes/skills/` does.
## Rollback
The pre-change configuration is preserved at:
`/home/hermes/.hermes/backups/config.yaml.20260725-before-tool-search`
Restore that file only if Tool Search causes a confirmed regression, validate
the YAML, and restart `hermes-gateway.service`.
## Related
- [[systems/hermes-agent]]
- [[systems/skills-index]]
- [[concepts/tool-calling]]
- [[runbooks/gateway-resource-resilience]]