Turn Your Agents into a Self-Evolving Team

DeepVista CLI connects your coding agents to a shared knowledge base —
so they don’t just execute, but learn, remember, and build on past work.

⭐ If this direction resonates, consider starring the repo — it helps more people discover it.

Who this is for

• Builders working with coding agents (Claude Code, Cursor, OpenCode)
• People experimenting with agent workflows and skills
• Anyone trying to turn scattered interactions into compounding knowledge

What this enables

Modern coding agents are powerful — but each interaction is stateless and isolated.

DeepVista changes that.

• Agents remember past work
• Agents share context through a common knowledge base
• Agents build on previous decisions and insights

Over time, this turns isolated executions into a:

→ a self-evolving team of agents that compound knowledge over time

Why DeepVista CLI?

Most knowledge tools trap your data in a GUI. DeepVista CLI gives you full access to your knowledge base from the terminal — and lets your AI agents use it too.

• 🧠 Capture knowledge — notes, insights, decisions — as you work
• 🔍 Hybrid search — vector + keyword across everything you've saved
• 🤖 Agent-native — Claude Code, Cursor, OpenCode use it as a skill
• 📋 Run Skills — execute structured AI workflows from the command line
• 💬 Chat — talk to the DeepVista AI agent in your terminal
• 🖥️ Full TUI — four-panel terminal UI for chat, notes, skills, and memory

Related Ideas

This repo is part of a broader exploration of how agent skills are structured and executed.

If you're interested in:

• how different types of skills should be modeled
• how to handle stateless vs stateful execution
• how to make agent workflows more reliable

→ Read more: https://go.deepvista.ai/agent-skills-2d

Quick Start

One-line install

pip install deepvista-cli

Or with uv / pipx:

uv tool install deepvista-cli    # or: pipx install deepvista-cli

Login and go

deepvista auth login              # opens browser
deepvista notes +quick "My first note from the terminal"
deepvista card +search "founder mindset"
deepvista chat +send "What did I learn last week?"

That's it. Your knowledge base is now accessible from any terminal.

For AI Agents

  Claude Code      Cursor    OpenCode    and any agent that supports skills

Install once, then talk to your agent. The agent handles authentication and all commands on your behalf.

curl -sSL https://raw.githubusercontent.com/DeepVista-AI/deepvista-cli/main/install.sh | bash

The script auto-detects your package manager (uv, pipx, or pip) and copies the consolidated deepvista skill into your agent's skill directory (Claude Code, OpenCode, Cursor, OpenClaw). If you upgraded from an earlier release, it also sweeps out the 12 legacy deepvista-* skills so you don't end up with both.

Prefer GitHub's CLI? gh skill install DeepVista-AI/deepvista-cli works too (GitHub CLI ≥ 2.90, preview).

Get Started

Open your agent and paste:

Help me get started with DeepVista. Walk me through logging in.

Your agent will open the browser login page, guide you through pasting the auth code, and confirm you're logged in.

Claude Code tip: To skip skill file read confirmations:

claude config set allowedPaths "~/.claude/skills" --global

Auto-Capture

The install script enables auto-capture in every detected agent — no manual setup required.

Your agent silently saves facts, decisions, insights, and action items to your DeepVista knowledge base as you work. The install is idempotent — re-running won't duplicate anything.

Skill Reference

One skill, deepvista, with per-subcommand detail under skills/deepvista/reference/.

The agent loads only the index at first; when a subcommand is needed, it reads the matching reference file on demand.

Use Cases

🎧 Capture Insights from Podcasts

Build a searchable knowledge base from founder interviews:

I just listened to the Lenny's Podcast episode with Brian Chesky about founder mode.
Save this to my knowledge base as a note titled "Brian Chesky — Founder Mode".

🔬 Research and Synthesize

Once you've captured 10–20 notes, ask your agent to find patterns:

Search for cards about growth and early-stage execution,
then run my Research Synthesis skill focused on:
what separates high-growth founders, common 0→1 mistakes.

📘 Build a Founder Playbook

I have a new startup idea to evaluate. Idea: [your idea].
1. Search my knowledge base for any idea validation frameworks
2. Find my idea evaluation Skill
3. Run it with the above context

Export as a reusable, shareable skill:

Export my founder playbook Skill as a SKILL.md file so I can share it with my team.

More Skill Patterns

Terminal UI (TUI)

A full four-panel terminal interface — chat, notes, skills, and memory in one view.

pip install 'deepvista-cli[ui]'
deepvista ui

Commands

deepvista card       # Knowledge cards (all types)
deepvista skill      # Executable workflows
deepvista vistabase  # Implicit context from Chat (read-only)
deepvista chat       # Conversational AI agent
deepvista notes      # Quick note management (shorthand)

card — Knowledge cards

deepvista card list [--type TYPE] [--status pinned|archived|normal] [--limit N]
deepvista card get <card_id>
deepvista card create --type TYPE --title "Title" [--content "..."] [--tags '["t1"]']
deepvista card update <card_id> [--title "..."] [--content "..."]
deepvista card delete <card_id>
deepvista card +search "query text" [--type TYPE] [--limit 10]
deepvista card +similar <card_id>
deepvista card +pin <card_id>
deepvista card +archive <card_id>

Card types: person · organization · message · todo · topic · keypoint · file · note · skill · skill_run

skill — Executable workflows

deepvista skill list [--limit N]
deepvista skill get <skill_id>
deepvista skill run <skill_id> [--input "context"]     # streams NDJSON
deepvista skill status <run_chat_id>

vistabase — Implicit context

Automatically accumulated from Chat. Read-only — updates happen through conversation.

deepvista vistabase show [--limit N]
deepvista vistabase search "query text" [--limit N]

deepvista memory is a deprecated alias that works identically.

chat — AI agent

deepvista chat sessions [--limit N] [--offset N] [--search "query"]
deepvista chat get <chat_id>
deepvista chat delete <chat_id>
deepvista chat +send "your message" [--chat-id ID] [--new]   # streams NDJSON

notes — Quick note management

Convenience alias for card commands filtered to type=note.

deepvista notes list [--limit N]
deepvista notes get <note_id>
deepvista notes create --title "Title" [--content "..."]
deepvista notes update <note_id> [--title "..."] [--content "..."]
deepvista notes delete <note_id>
deepvista notes +quick "Quick note from a single line of text"

Install Options

From PyPI

pip install deepvista-cli
pipx install deepvista-cli
uv tool install deepvista-cli

With TUI support

pip install 'deepvista-cli[ui]'
uv tool install 'deepvista-cli[ui]'

From GitHub

pip install git+https://github.com/DeepVista-AI/deepvista-cli.git
uv tool install git+https://github.com/DeepVista-AI/deepvista-cli.git

As a Claude Code plugin

The plugin is additive on top of the CLI — it wires the DeepVista skill catalog into Claude Code on every SessionStart. Install in this order:

# 1. Install the CLI (required — the plugin calls it)
uv tool install deepvista-cli    # or: pip install deepvista-cli
deepvista auth login

# 2. Then, inside Claude Code:
#    /plugin marketplace add DeepVista-AI/deepvista-cli
#    /plugin install deepvista@deepvista-ai

Without the CLI on PATH, the plugin's SessionStart hook exits silently — you'll see no errors but no skills either.

For development

uv sync
uv pip install -e '.[ui]'
uv run deepvista --help

Uninstall

curl -sSL https://raw.githubusercontent.com/DeepVista-AI/deepvista-cli/main/uninstall.sh | bash

Removes the CLI, all skills from agent directories, and auto-capture blocks from config files.

Authentication

deepvista auth login                   # opens browser (default)
deepvista auth login --code XXXX-XXXX  # headless / non-interactive
deepvista auth status                  # check current auth
deepvista auth logout                  # clear tokens

For headless environments, visit the web app at /cli, sign in, and paste the code.

Profiles

deepvista config set local   --api-url http://localhost:8080
deepvista config set staging --api-url https://api-staging.deepvista.ai
deepvista --profile local card list
deepvista config list

Resolution order (first wins): CLI flags → named profile → built-in default

Global Flags

Global flags must come before the resource name: deepvista --profile local card list

Reference

deepvista-cli/
├── deepvista_cli/
│   ├── main.py              # CLI entry point
│   ├── config.py            # Config + profiles
│   ├── auth/                # Login, token storage, callback server
│   ├── client/              # HTTP client, SSE streaming
│   ├── commands/            # card, skill, memory, chat, notes, auth, config
│   ├── tui/                 # Terminal UI (requires [ui] extra)
│   └── output/              # JSON + table formatters
└── skills/                  # SKILL.md files for agent integration

🤝 Contributing

We welcome contributions! Whether it's bug reports, feature requests, documentation updates, or code — open an issue or submit a PR.

👥 Community

• 🌟 Star us on GitHub — helps us grow
• 🐛 Report issues
• 📦 PyPI package
• 🌐 DeepVista Web App

Support Us

If you find DeepVista CLI useful, please ⭐ star this repo — it helps more people discover it!

License

DeepVista CLI is open source. See LICENSE for details.