Hermes Tool Slimmer
aliasocracy/hermes-tool-slimmer
Hermes Agent plugin that sends only the relevant tool schemas with each model request
Hermes Tool Slimmer is a Hermes Agent plugin that trims the tool schemas sent with each model request, so the model sees your always-on tools plus the best matches for the current prompt instead of the whole catalog. It includes a dashboard that shows how many tokens it saves.
What Hermes Tool Slimmer does
Every Hermes model call carries the JSON schema of every enabled tool, which the README puts at roughly 15 to 20k tokens per request on installs with dozens of native and MCP tools. Tool Slimmer ranks tools against the current prompt with BM25, using boosts for name, toolset, parameters and aliases, then sends the always-on tools plus the top matches. The default keyword mode is fully local, with no embeddings, extra model calls or network access.
On a 38-tool Hermes install, averaged over 40 prompts, the README reports about 15.1k tool tokens per request without slimming, about 7.9k in keyword mode and about 4.3k in the experimental jev mode. The selector fails open by sending the original tool list on any error, and a built-in tool_slimmer_request_full_tools fallback lets the model ask for the full catalog on the next request. On Hermes v0.19 and later it hooks in through the built-in llm_request middleware, so no core patch is needed.
Key features
- BM25 keyword ranking with name, toolset, parameter and alias boosts
- Fails open to the original tool list, plus a full-catalog fallback tool
- Separate profiles and include or exclude lists for Telegram, Slack, CLI, cron and webhooks
- Dashboard with savings, recent decisions, guided setup and a tool index browser
- A dry_run setting to watch decisions before changing any request
- Leaves approval prompts, disabled tools, auth and safety policy untouched
When to use it
- Cutting per-request token cost on a Hermes install with dozens of native and MCP tools
- Giving messaging, cron and CLI entry points different tool budgets
- Checking in dry-run mode how much a trimmed tool list would save before enabling it
Who it is for: Hermes Agent users running many native or MCP tools who want to cut the tool-schema tokens sent on every model call.
How it fits with Hermes Agent
It is a Hermes Agent plugin that hooks into the llm_request middleware on Hermes v0.19 and later, and it adds a hermes tool-slimmer command and a dashboard plugin.
How to install Hermes Tool Slimmer
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
git clone https://github.com/aliasocracy/hermes-tool-slimmer.git "$HOME/hermes-tool-slimmer"
bash "$HOME/hermes-tool-slimmer/scripts/install-hermes-tool-slimmer.sh"
hermes tool-slimmer doctorRequirements: A Hermes Agent install on the same machine; Hermes v0.19 or later needs no core patch
Note: The jev selector mode is labelled experimental in the README.
FAQ
What is Hermes Tool Slimmer?
Hermes Tool Slimmer is a Hermes Agent plugin that reduces tool-schema overhead by sending only the tools relevant to the current turn. It ranks tools against your prompt and sends the always-on tools plus the best matches.
How do I install Hermes Tool Slimmer?
Clone the repository, run scripts/install-hermes-tool-slimmer.sh, then check the result with hermes tool-slimmer doctor. The installer installs the package, copies the dashboard plugin, enables the plugin in Hermes and restarts services. On Hermes builds with dashboard plugin install, you can instead paste aliasocracy/hermes-tool-slimmer on the Plugins page.
Does Hermes Tool Slimmer change which tools are allowed?
No. It only selects which tool schemas are sent with a request. Approval prompts, disabled tools and toolsets, auth and runtime safety policy are untouched.
Similar plugins for Hermes Agent
All pluginsHermes plugin that runs browser tasks on the Obscura headless browser instead of Chrome
Elevatormusic Hermes Warm CompactionHermes Agent context engine where the main model writes the compaction handoff on its cached prompt prefix
Bergschloss Refine Cycle for Hermes AgentSelf-improvement plugin for Hermes Agent: spots repeated failures, saves the smallest skill or memory fix
willingning-coder Eagle EyeFive-layer skill retrieval plugin for Hermes Agent: triggers, FTS5, synonyms, embeddings and RRF fusion
Xquik-dev Hermes TweetHermes Agent plugin for X (Twitter) search, monitoring, follower exports and approval-gated actions
eas4ai Hermes Context ManagerSilent context compression plugin for Hermes Agent that shortens tool output and removes repeated work
Related guides: How to install Hermes Agent · How to run Hermes Agent securely