AI Agent Runtime

Sovereign Hermes. An autonomous AI agent with 50+ built-in tools.

Model-agnostic agent runtime with terminal access, web search, browser automation, persistent memory, and multi-agent orchestration. Every instruction executed on UK sovereign hardware.

Request Access →
REST API + MCP protocol. Deploy in minutes.
50+ built-in tools
Multi-agent orchestration
Persistent memory
UK sovereign hardware
Built-in Tools

Everything an agent needs. Nothing it does not.

Hermes ships with 50+ tools that let it interact with terminals, files, the web, browsers, code, images, and its own memory. No plugins to install.

Terminal execution

Run shell commands in a sandboxed environment. Install packages, run scripts, pipe outputs.

📁

File management

Read, write, search, and organise files. Parse PDFs, CSVs, DOCX, and structured data.

🔍

Web search

Private web search via self-hosted SearXNG. No queries leaked to third-party APIs.

🌐

Browser automation

Navigate pages, extract data, fill forms, take screenshots. Headless Chromium on sovereign infrastructure.

💻

Code execution

Run Python, JavaScript, bash, and more. Analyse data, generate visualisations, test logic.

👁

Vision

Analyse images, screenshots, charts, and documents. Extract text, describe content, identify patterns.

🧠

Persistent memory

Store and recall facts across sessions via PostgreSQL and ChromaDB semantic search.

🤝

Agent delegation

Spawn specialised sub-agents for complex tasks. Coordinate multi-step workflows autonomously.

Architecture

Three ways to talk to Hermes. One sovereign runtime.

Access the agent through a REST API, server-sent events for streaming, or the MCP protocol for tool-native integration. All routes execute on the same UK hardware.

Access Methods

  • REST API with JSON request/response
  • Streaming via Server-Sent Events (SSE)
  • Async task queue with polling
  • MCP protocol for tool ecosystems

Infrastructure

  • UK-owned GPU compute
  • PostgreSQL persistent memory
  • ChromaDB vector embeddings
  • Sandboxed execution containers
Execution Modes

Synchronous, streaming, or fire-and-forget.

Choose the execution model that fits your workflow. All three modes are available on every endpoint.

Synchronous

Send a request, wait for the complete response. Ideal for simple queries and integrations.

POST /v1/agent/run

Streaming (SSE)

Receive tool calls and reasoning in real time as the agent works through a task.

POST /v1/agent/stream

Async Tasks

Submit long-running tasks and poll for results. Background processing for complex workflows.

POST /v1/agent/task
Quick Start

One request. The agent handles the rest.

Send a natural language instruction. Hermes selects the right tools, executes them, and returns structured results.

# Send a task to Hermes
curl -X POST https://api.hush-ai.uk/v1/agent/run \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "instruction": "Search for the latest UK AI regulation updates, summarise the key points, and save to a report file.",
    "model": "auto",
    "tools": "all"
  }'
Multi-Agent Orchestration

One agent delegates. Specialists execute.

Hermes can spawn sub-agents with constrained tool sets and specific roles. A research agent searches and summarises. A code agent writes and tests. A review agent validates. The orchestrator coordinates them all.

📖

Research Agent

Searches the web, reads documents, extracts facts, and produces structured summaries.

Code Agent

Writes, tests, and debugs code. Runs in a sandboxed environment with full terminal access.

📋

Review Agent

Validates outputs against criteria. Checks facts, verifies formatting, flags inconsistencies.

📦

Data Agent

Extracts, transforms, and loads data. Parses files, queries APIs, builds datasets.

Persistent Memory

Session continuity that actually works.

Hermes remembers context across conversations. Facts, preferences, and project state are stored in PostgreSQL with semantic recall powered by ChromaDB vector embeddings. Your agent picks up exactly where it left off.

Structured Storage

  • PostgreSQL for relational facts
  • Session history and turn logs
  • User preferences and project state
  • Exportable and auditable

Semantic Search

  • ChromaDB vector embeddings
  • Recall by meaning, not keywords
  • Cross-session context retrieval
  • Automatic relevance ranking
Use Cases

From compliance reports to code reviews.

Hermes handles multi-step workflows that would take a human hours. Here are some examples.

Compliance report generation

Gather regulatory data, cross-reference requirements, draft structured compliance reports with citations.

Research + File + Memory

Literature review

Search academic sources, extract key findings, synthesise into a structured review with references.

Web Search + Browser + File

Data extraction pipelines

Scrape websites, parse documents, transform data, and output clean structured datasets.

Browser + Code + File

Customer onboarding

Process applications, verify documents, generate welcome packs, update CRM records automatically.

File + Code + Memory

Document processing

Ingest PDFs, extract key fields, classify documents, route to appropriate workflows.

Vision + File + Code

Code review automation

Analyse pull requests, identify bugs, check style compliance, generate review comments.

Code + Terminal + File

Infrastructure monitoring

Check service health, parse logs, identify anomalies, generate alerts and incident summaries.

Terminal + Web + Memory
Getting Started

From API key to autonomous agent in four steps.

01

Get API key

Sign up and receive your API credentials. Takes under two minutes.

02

Send first task

POST a natural language instruction to the agent endpoint. Hermes selects tools automatically.

03

Review execution

Watch the agent work via streaming. Every tool call is logged and auditable.

04

Deploy workflows

Chain tasks, configure memory, set up multi-agent pipelines. Scale as needed.

Data Sovereignty

Your data never leaves UK soil.

Every tool call, every file operation, every web search, every memory write executes on hardware we physically own in the United Kingdom. No US cloud. No CLOUD Act exposure.

🇬🇧

UK Hardware

All compute on UK-owned, UK-located servers. Not rented cloud.

🛡️

CLOUD Act Immune

UK company, UK hardware. US law does not apply.

🔒

Sandboxed Execution

Each agent session runs in an isolated container. No cross-tenant leakage.

📜

Full Audit Trail

Every tool call logged with timestamp. Exportable compliance records.

🚫

Zero Retention

Session data purged on completion unless you enable persistent memory.

⚖️

ICO Registered

Data controller C1912355. UK GDPR compliant. DPA available on request.

Deploy an autonomous agent on sovereign infrastructure.

50+ tools. Multi-agent orchestration. Persistent memory. All on UK hardware you can audit.

Request Access →

Or email [email protected] directly.