Skip to main content

CIRI — Contextual Intelligence and Reasoning Interface

Ciri is an open-source, autonomous AI copilot that lives inside your workspace. She is not a coding assistant — she is a general-purpose intelligence that adapts to whatever domain you work in: software engineering, accounting, marketing, branding, sales, legal research, content production, data analysis, or any discipline you train her on.

Ciri builds a deep, persistent understanding of your workspace, connects to your real browser, executes sandboxed scripts, and permanently expands her own capabilities by creating new Skills, Toolkits, and SubAgents — all from a single terminal interface.


Why Ciri?

FeatureCiriGeneric AI Chat
Persistent workspace memory✅ Two-level (global + project)❌ Stateless
Domain-agnostic training✅ Any field: dev, finance, marketing…❌ General only
Self-evolving capabilities✅ Builds her own tools❌ Fixed toolset
Real browser integration (CDP)✅ Your sessions & cookies❌ Scrapers only
Human-in-the-loop approvals✅ Per-action, editable❌ All-or-nothing
MCP Toolkit ecosystem✅ Any standard MCP server
Local-first, private history✅ SQLite on-disk❌ Cloud
Multi-provider LLM✅ OpenAI, Anthropic, OpenRouter…Varies

What Can Ciri Do?

Ciri's capabilities are defined by the workspace she inhabits and the training she receives via /sync. Examples across domains:

Software Engineering: Read codebases, write and test code, manage git, create PRs, run CI pipelines, refactor, document.

Business Operations: Draft proposals, analyze financials, automate reporting, manage project trackers, generate slide decks.

Marketing & Branding: Research competitors, draft copy, generate visuals, schedule campaigns, analyze engagement data.

Sales & CRM: Qualify leads, draft outreach, summarize call notes, update CRM via MCP toolkits, track pipeline.

Data & Research: Crawl the web for primary sources using your authenticated browser, summarize papers, build data pipelines.

Document Work: Fill PDF forms, process DOCX/PPTX/XLSX files, extract structured data, produce formatted reports.

The key insight: Ciri trains herself to match your domain. Run /sync in a marketing agency workspace and she learns your brand voice, tools, and workflows. Run it in a fintech codebase and she learns your stack, conventions, and release process.


System Architecture


Key Concepts

Skills

Reusable capability packages that give Ciri domain knowledge and executable tools. A skill is a directory containing a SKILL.md (playbook) and a scripts/ folder of Python helpers. Skills live in the core harness (~/.local/share/ciri/skills/) for global availability or your project harness (.ciri/skills/) for workspace-specific ones. Ciri ships with built-in skills for PDF forms, Office documents, web testing, MCP server building, and more.

Skills Guide

Toolkits (MCP)

Ciri's ToolkitInjectionMiddleware is an MCP client that connects to any standard Model Context Protocol server. Ask Ciri to "build a toolkit for Slack" or "connect to my PostgreSQL database" and the toolkit_builder subagent generates an MCP server, tests it, and hot-reloads it — no restart required.

Toolkits Guide

SubAgents

Specialized agents with scoped system prompts and tool access. The five built-in subagents (web_researcher, skill_builder, toolkit_builder, subagent_builder, trainer_agent) handle complex delegated tasks. You can define your own in .ciri/subagents/ using YAML configuration.

SubAgents Guide

Core Harness

A two-level filesystem that separates global capabilities (available across all your projects) from project-specific ones. The global harness lives in ~/.local/share/ciri/ (Linux) or equivalent OS data dir, and the project harness in .ciri/ at your workspace root.

Core Harness Architecture

Self-Evolution

Run /sync and the Trainer Agent audits your environment, identifies capability gaps, and orchestrates builder subagents to create new Skills, Toolkits, or SubAgents — then hot-reloads them without a restart. The more you use Ciri in a domain, the more specialized she becomes.

Self-Evolution & Training


Quick Start

# Install
pip install ciri-copilot
# or with uv (recommended)
uv pip install ciri-copilot

# Run
ciri

# On first run, Ciri walks you through model selection.
# Then analyze your workspace:
/sync

Full Getting Started Guide


SectionWhat's Inside
Getting StartedInstallation, first run, model configuration
CLI ReferenceAll /commands, @triggers, keyboard shortcuts
Skills GuideCreating and using domain capability packages
Toolkits GuideMCP server integration
SubAgents GuideDelegation and custom agent roles
ArchitectureHow the graph, middleware, and harness fit together
FeaturesMemory, web research, HITL, autocomplete, script executor
Built-in SkillsPDF, Office docs, web testing, and more
InternalsGraph execution, middlewares, self-evolution deep dive
Use CasesReal-world walkthroughs across domains