Humanlike: developer tool for Hermes Agent

Deterministic persona, context, memory and privacy controls for chat agents, with a Hermes plugin

What Humanlike does

Humanlike classifies each user turn, selects a social and cognitive mode, assembles bounded guidance with mandatory truth and privacy rules, and returns metadata. The host keeps responsibility for model calls, tools, transport, transcripts and policy. The core runtime uses only the Python standard library and does not call an LLM or the network. Routing supports Russian and English. Persona anchoring, repetition control, calibrated stance and drift signals are included, and optional SQLite memory stays off unless the host enables it with explicit consent settings.

A reference Hermes directory plugin registers pre_llm_call, transform_llm_output, post_llm_call and on_session_finalize hooks. The README says to install the package into Hermes' own Python environment, then run hermes plugins enable humanlike-agent-kit --no-allow-tool-override, and to start a new Hermes session afterwards. The CLI commands route, doctor and eval print JSON, and a 40-case offline conformance suite covers routing, privacy, context budgets, policy, disclosure, stance, memory and drift.

Key features

Deterministic Russian and English routing across cognitive modes and social moves
Bounded context plans with mandatory truth and privacy guidance
Persona anchoring, repetition control, calibrated stance and drift signals
Optional evidence-aware SQLite memory, off by default
Reference Hermes plugin with pre_llm_call, transform_llm_output, post_llm_call and on_session_finalize hooks
40-case offline conformance suite run with humanlike eval

When to use it

Keeping a Hermes persona consistent from SOUL.md across many turns without extra model calls
Limiting how much context and memory is injected into each prompt
Checking routing and privacy behavior offline before deploying an agent

Who it is for: Developers building conversational agents, including Hermes Agent profiles, who want predictable and auditable behavior rules.

How it fits with Hermes Agent

A provider-neutral library that includes a reference Hermes plugin, enabled with hermes plugins enable humanlike-agent-kit.

How to install Humanlike

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/AlekseiUL/humanlike.git
cd humanlike
python -m pip install .
humanlike eval

Requirements: Git and Python 3.11 or newer; Linux and macOS support the full stack, and native Windows supports the core runtime, CLI and memory-off Hermes plugin

Note: The README marks this as a public beta (0.1.2) whose API and configuration schema may change before 1.0.

FAQ

What is Humanlike?

Humanlike is a deterministic behavior-planning layer for conversational AI agents. It classifies each turn, picks a mode, builds bounded guidance and returns privacy-aware metadata without calling an LLM.

Does Humanlike work with Hermes Agent?

Yes, through a reference Hermes directory plugin named humanlike-agent-kit. Install the package into Hermes' Python environment, enable the plugin, and start a new session.

What do I need to run Humanlike?

You need Git and Python 3.11 or newer. The installed runtime has no third-party Python dependencies, and the optional SQLite memory ledger is POSIX-only in 0.1.x.