Availability: the independent Cognivoy package has not yet completed its first public PyPI release. Use the source installation below until a Cognivoy release appears on PyPI.

Full local Web app + CLI, no clone required. Needs Python 3.11–3.13 and a Node.js 20+ runtime on PATH (the packaged Next.js standalone server is spawned by cognivoy start).

mkdir -p my-cognivoy && cd my-cognivoy
pip install -U cognivoy
cognivoy init     # prompts for ports + LLM provider + optional embedding
cognivoy start    # starts backend + frontend; keep the terminal open

cognivoy init prompts for backend port (default 8001), frontend port (default 3782), LLM provider / base URL / API key / model, and an optional embedding provider for Knowledge Base / RAG.

After cognivoy start, open the frontend URL printed in the terminal — by default http://127.0.0.1:3782. Press Ctrl+C in that terminal to stop both backend and frontend. Skipping cognivoy init is fine for a quick trial; the app boots with default ports and empty model settings, configure them later in Settings → Models.

For development against a checkout. Use Python 3.11–3.13 and Node.js 22 LTS to match CI and Docker.

git clone https://github.com/chenchunrun/Cognivoy.git cognivoy
cd cognivoy

# Create a venv (macOS/Linux). Windows PowerShell:
#   py -3.11 -m venv .venv ; .\.venv\Scripts\Activate.ps1
python3 -m venv .venv && source .venv/bin/activate
python -m pip install --upgrade pip

# Install backend + frontend deps
python -m pip install -e .
( cd web && npm ci --legacy-peer-deps )

cognivoy init
cognivoy start --dev

cognivoy start builds the local web/ frontend for production once and reuses it; --dev runs Next.js with HMR. Config layout, ports, and Ctrl+C match Option 1.

conda create -n cognivoy python=3.11
conda activate cognivoy
python -m pip install --upgrade pip

pip install -e ".[dev]"             # tests/lint tools
# Local vector/hybrid retrieval and Docling are temporarily unavailable.
# Keyword retrieval is included; existing knowledge-base data is preserved.
pip install -e ".[partners]"        # Partner IM channel SDKs + MCP client
pip install -e ".[matrix]"          # Matrix channel without E2EE/libolm
pip install -e ".[matrix-e2e]"      # Matrix E2EE; requires libolm
pip install -e ".[math-animator]"   # Manim addon; requires LaTeX/ffmpeg/system libs

rm -f web/.next/dev/lock web/.next/lock
cognivoy start --dev

One container for the full Web app. The following GitHub Container Registry images become available after the first public Cognivoy release:

• ghcr.io/chenchunrun/cognivoy:latest — stable release
• ghcr.io/chenchunrun/cognivoy:pre — pre-release, when available

See CONTAINERIZATION.md for podman/rootless/read-only-rootfs deployments and the full per-installation guide.

docker run --rm --name cognivoy \
  -p 127.0.0.1:3782:3782 \
  -v cognivoy-data:/app/data \
  ghcr.io/chenchunrun/cognivoy:latest

Only 3782 needs to be published. The browser talks exclusively to the frontend origin; the Next.js middleware (web/proxy.ts) forwards /api/* and /ws/* to the FastAPI backend inside the container. Publishing 8001 (-p 127.0.0.1:8001:8001) is optional — handy only for hitting the API directly with curl or scripts.

Open http://127.0.0.1:3782. The container creates /app/data/user/settings/*.json on first boot; configure model providers from the Web Settings page. Config, API keys, logs, workspace files, memory, and knowledge bases persist in the cognivoy-data volume.

• Different host ports: change the left side of each -p host:container mapping (e.g. -p 127.0.0.1:8088:3782). If you change container-side ports in /app/data/user/settings/system.json, restart and update the right side of each mapping to match.
• Detached: add -d, then docker logs -f cognivoy to follow, docker stop cognivoy to stop, docker rm cognivoy before reusing the name. The cognivoy-data volume keeps your settings and workspace across restarts.

Remote Docker / reverse proxy: the browser only talks to the frontend origin (:3782); the in-container Next.js middleware forwards /api/* and /ws/* to the backend server-side. For the common single-container case you don't configure an API base at all — just point your reverse proxy / TLS terminator at :3782. You only need an API base for a split deployment (backend in a separate container/host): set next_public_api_base in data/user/settings/system.json to the in-network address the frontend server uses to reach the backend (it's read server-side, never sent to the browser).

{
  "next_public_api_base": "http://backend:8001"
}

next_public_api_base_external (and its alias public_api_base) are accepted as lower-precedence fallbacks. CORS uses frontend origins, not API URLs. With auth disabled, Cognivoy permits normal HTTP/HTTPS browser origins by default. With auth enabled, add exact frontend origins:

{
  "cors_origins": ["https://cognivoy.example.com"]
}

docker run --rm --name cognivoy \
  -p 127.0.0.1:3782:3782 -p 127.0.0.1:8001:8001 \
  --add-host=host.docker.internal:host-gateway \
  -v cognivoy-data:/app/data \
  ghcr.io/chenchunrun/cognivoy:latest

When you don't need the Web UI. The CLI-only package is installed from a source checkout, not from PyPI.

git clone https://github.com/chenchunrun/Cognivoy.git
cd cognivoy

# Create a venv (macOS/Linux). Windows PowerShell:
#   py -3.11 -m venv .venv-cli ; .\.venv-cli\Scripts\Activate.ps1
python3 -m venv .venv-cli && source .venv-cli/bin/activate
python -m pip install --upgrade pip

python -m pip install -e ./packaging/cognivoy-cli
cognivoy init --cli
cognivoy chat

cognivoy init --cli shares the same data/user/settings/ layout as the full app but skips the backend/frontend port prompts and defaults embeddings to off (choose Yes for vector/hybrid retrieval; keyword retrieval needs no embedding profile). It still writes a complete runtime layout (system.json, auth.json, integrations.json, model_catalog.json, main.yaml, agents.yaml) and still prompts for the active LLM provider and model.

cognivoy chat                                          # interactive REPL
cognivoy chat --capability deep_solve --tool rag --kb my-kb
cognivoy run chat "Explain Fourier transform"
cognivoy run deep_solve "Solve x^2 = 4" --tool rag --kb my-kb
cognivoy kb create my-kb --doc textbook.pdf
cognivoy memory show
cognivoy config show

The local cognivoy-cli install ships no Web assets or server dependencies. Keep the source checkout around — the editable install points to it. To add the Web app later, install the PyPI package (Option 1) and run cognivoy init + cognivoy start from the same workspace.

The built-in office skills — docx / pdf / pptx / xlsx — work by having the model write a short Python script (python-docx, reportlab, openpyxl, …), run it through the exec / code_execution tools, and hand back a download URL. Those tools mount only when a backend provides OS-enforced isolation for the current task directory:

• Linux with bubblewrap: each task receives a separate mount namespace and network namespace, so code execution can be enabled.
• Restricted subprocess and the legacy shared-workspace runner: rejected by default because they cannot prevent one task from reading another task's files. The runner endpoint is additionally bearer-authenticated and placed on an internal Docker network.

sandbox_allow_subprocess remains a compatibility setting, but does not bypass the per-task isolation requirement. If no qualifying backend is available, exec and code_execution stay disabled and Office skills cannot generate files.

Everything under data/user/settings/ is plain JSON/YAML. The Settings page in the browser is the recommended editor.

Project-root .env is not read as an application config file. For a minimal model setup, open Settings → Models, add an LLM profile (Base URL / API key / model name), and save. LlamaIndex knowledge retrieval defaults to model-free keyword mode for new installations; add an embedding profile only for vector/hybrid retrieval. Existing saved vector/hybrid choices remain unchanged. See model-free knowledge retrieval for configuration, migration, and matching limits, and optional capabilities for animation/media deployment requirements.

Chat is the default capability and where most work begins. A single thread can talk normally, call tools, ground itself in selected knowledge bases, read attachments, generate images, consult subagents, write notebook records, and continue with the same context across turns.

The loop is deliberately simple: the model thinks in rounds, calls tools when useful, observes the results, and finishes with a tool-free message. ask_user is special — instead of guessing, the agent can pause the turn, ask a structured clarifying question, and resume once you answer.

User-toggleable tools are brainstorm, web_search, paper_search, reason, and geogebra_analysis — plus imagegen and videogen once you configure the matching generation model. Contextual tools such as rag, kb_files, read_source, read_memory, write_memory, read_skill, load_tools, exec, web_fetch, ask_user, list_notebook, write_note, github, and consult_subagent mount automatically when the turn has the right context.

Context comes in two kinds: sticky session context (subagent, knowledge bases, persona, model, voice) lives on the composer toolbar and persists across turns; one-time references (files, chat history, books, notebooks, question bank, imported agents) come from the + menu for a single turn.

Chat is also the launch point for deeper capabilities: Quiz for question generation, Research for cited reports, Visualize for charts / diagrams / animations, and — under More Capabilities — Solve for worked reasoning and Mastery Path for learning-plan flows.

Partners are persistent companions with their own soul, model policy, library, memory, and channels. They are not a separate bot engine: every inbound web or IM message becomes a normal ChatOrchestrator turn inside a partner-scoped workspace. A partner is "a chat that has a personality and a phone number."

Each partner has a SOUL.md, model selection, channels, tool policy, and assigned library. Knowledge bases, skills, and notebooks are copied into data/partners/<id>/workspace/, so the same RAG, skill, notebook, and memory tools work without special cases. A partner reads its owner's memory but writes only its own.

The channel layer is schema-driven and can connect to IM platforms such as Feishu, Telegram, Slack, Discord, DingTalk, QQ/NapCat, WeCom, WhatsApp, Zulip, Mattermost, Matrix, Mochat, and Microsoft Teams depending on installed extras and configured credentials. A partner can also be connected as a subagent and consulted from a normal chat turn — see My Agents below.

My Agents turns other agents into context for Cognivoy, and does two distinct things. Connect a live agent — a Claude Code, Codex, Gemini, Kimi, opencode, or MiMo Code CLI on your machine, or one of your Partners — and consult it from inside a chat turn: Cognivoy actually runs the other agent and streams its work into the Activity panel via the consult_subagent tool. Select it with the Agent chip (or type @), and set how many rounds the consult may take.

Import past conversations — bring in your existing Claude Code and Codex history as named, searchable, resumable agents. Pick which days to import; refreshing re-syncs them. Reference an imported conversation from any chat turn via + → My Agents, and Cognivoy reads it as a third-party transcript — it stays their conversation, not Cognivoy's own voice.

Co-Writer is a split-view Markdown workspace for reports, tutorials, notes, and long-form learning artifacts. Documents autosave and render a live preview (KaTeX math, diagram fences), and can be saved back into notebooks when a draft becomes reusable context.

Its defining idea is surgical editing: select a span and ask Cognivoy to rewrite, expand, or shorten it. The edit agent can ground the change in a knowledge base or web evidence, keeps a trace of its tool calls, and shows every change as an accept/reject diff — so nothing lands until you approve it.

Book turns selected sources into an interactive living book — not a static PDF, but a reading environment built from typed blocks. A book can start from knowledge bases, notebooks, question banks, or chat history; the creation flow proposes a chapter outline before content is generated, so you review the shape instead of accepting a blind one-shot output.

Each chapter compiles into typed blocks — text, callouts, quizzes, flash cards, timelines, code, figures, interactive HTML, animations, concept graphs, deep dives, and user notes — and every page has its own Page Chat. Blocks are editable: insert, move, regenerate, or switch a block's type without rewriting the chapter. Maintenance commands such as cognivoy book health and cognivoy book refresh-fingerprints help detect when source knowledge has drifted from compiled pages.

Knowledge bases are the document collections behind RAG — they ground Chat turns, Co-Writer edits, Book generation, and Partner conversations. What's distinctive is a choice of retrieval engines: LlamaIndex (the default engine, with model-free BM25 and optional vector/hybrid retrieval), PageIndex (hosted, reasoning retrieval with page-level citations), GraphRAG and LightRAG (knowledge-graph retrieval), LightRAG Server (retrieval offloaded to an external LightRAG instance you connect over HTTP), Tencent IMA (a library you curate in IMA, searched over its OpenAPI), or a linked Obsidian vault the tutor reads and writes in place. Each KB is bound to one engine.

For the recommended lightweight setup, choose Settings → Embedding → Knowledge retrieval mode → Keyword (no model). Keyword/BM25 retrieval reuses extracted text without Ollama, embedding API calls, or a GPU. It does not promise cross-language semantic matching; scanned images still need text extraction.

Creating a KB, you either create new (upload documents and build a fresh index) or link existing (reuse an index built elsewhere, read in place with no re-index). Vector/hybrid re-indexing writes a new flat version-N directory and keeps prior ones; keyword indexing publishes its separate text sidecar atomically. A single document can be removed even from an error-state base — dropping a file that failed to parse without a full delete-and-rebuild. Document parsing — Text-only, MinerU, Docling, markitdown, or PyMuPDF4LLM — is chosen in Settings → Knowledge Base, with local model downloads off by default. The CLI mirrors the lifecycle with cognivoy kb list, info, create, add, search, set-default, and delete.

Learning Space is the library and personalization layer — where the things that persist live. Conversations & Materials holds your chat history, notebooks, and a question bank (each saved question keeps your answer, the reference answer, and an explanation). Personalization holds mastery paths, personas (behavior presets such as peer, research-assistant, teacher), skills (SKILL.md playbooks the model reads on demand), MCP Services — a curated store of hosted MCP servers you install for yourself in one click, plus any remote server you configure by URL — and CLI Apps, command-line tools from the CLI-Anything catalog that the chat agent calls directly, with each app's own usage guide loaded on demand. Everything here can be reused from Chat, Partners, Co-Writer, and Book.

You don't have to write every skill yourself — Import from EduHub browses the community catalog and downloads a skill straight into your library through a security gate (see Ecosystem).

Memory is a file-backed, three-layer system you can read, curate, and audit — deliberately not a hidden vector store. L1 is the workspace mirror plus an append-only event trace (trace/<surface>/<date>.jsonl); L2 is per-surface curated facts (L2/<surface>.md); L3 is cross-surface synthesis (L3/<profile|recent|scope|preferences>.md). Because L2 cites L1 and L3 cites L2, nothing in your profile is unaccountable.

The Memory Graph shows the whole pyramid — L3 synthesis at the centre, L2 in the middle ring, L1 traces on the outside — so you can trace any synthesized claim back to the exact raw event behind it. Memory is tracked across chat, notebook, quiz, kb, book, partner, and cowriter surfaces; the consolidator's Update / Audit / Dedup budgets are tuned in Settings → Memory.

Settings is the operational control plane, with a live status strip (backend health and resident memory across the process tree) and one card per area: Appearance (theme, interface and model output language, code-block styling), Network (API base, ports, CORS), Models (LLM, Embedding, Search, Text-to-Speech, Speech-to-Text, Image Generation, Video Generation), Knowledge Base (document parsing engine), Chat (tools, per-capability parameters, attachment caps), Partners & Agents (the subagents you can consult from a turn), and Memory (the consolidator's budgets).

Most sections use a draft-and-apply flow, so you can test a provider before committing it. Four themes ship in the box — Default, Cream, Dark, and Glass. Project-root .env files are intentionally ignored; runtime configuration lives under data/user/settings/*.json unless COGNIVOY_HOME or cognivoy start --home points the app elsewhere.

OpenAI Codex OAuth (experimental). Picking OpenAI Codex under Models → LLM replaces the API-key fields with a browser sign-in that runs against your own ChatGPT plan, so no OPENAI_API_KEY is needed. Tokens live only in data/system/user-secrets/<owner>/private/openai-codex/ — in the multi-container Compose deployment, outside every tree the exec sandbox can reach — and Cognivoy never reads or modifies your ~/.codex CLI login. The model list comes from that account's live catalog; signing in publishes the profile but only becomes the active model when no LLM is configured yet. Because a token authorizes one person's plan, the profile is not shareable through user grants — each account signs in for itself, ordinary users included: their card sits under Models → LLM, and the resulting models, catalog, and sign-out stay private to that account.

Default local Docker and Podman deployments use separate loopback networks and need a temporary bridge during sign-in. Follow the temporary local Codex OAuth bridge guide for the exact Docker, Compose, Podman, and teardown commands.

For a remote deployment, the browser's localhost and the server's localhost are different machines, so an ordinary reverse proxy alone cannot carry the browser's localhost callback to the server. Use an SSH tunnel as the callback bridge. The tunnel reaches the already-published Web port; Next.js rewrites only the exact callback path to the public callback broker, and the broker validates state before routing to the original OAuth operation. The callback listener remains on the backend loopback, ports 1455 and 1457 are not published, and this path supports the default Docker bridge network.

ssh -N -L 1455:127.0.0.1:3782 <ssh-user>@<server-host>

If Cognivoy reports fallback callback port 1457, use:

ssh -N -L 1457:127.0.0.1:3782 <ssh-user>@<server-host>

Run only the one command that matches the actual callback port; never run both. 3782 is only the example Web port: it is the configured frontend/container port reported as callback_forward_port. That value does not guarantee that the same port is listening on the SSH host's 127.0.0.1. If Docker or Podman publishes a different host port, or a reverse proxy listens on a different port, replace only the right-hand target port (3782 above) with the Web port actually listening on the SSH host's 127.0.0.1; keep the left-hand callback port as 1455 or 1457. <server-host> is the SSH host whose loopback owns that listening port. If the browser URL names a reverse proxy or load balancer, replace it with the correct SSH frontend host.

The CLI prints the tunnel command and then immediately tries to open the browser. On a remote deployment, keep the authorization page open without completing it, establish the printed tunnel in another terminal, and only then continue authorization.

Remote-topology detection has a localhost boundary. If Web itself is reached through an SSH or IDE localhost forward, the browser cannot tell that the server is remote. For the current Web operation, leave its authorization page unfinished, read redirect_uri in that operation's authorize URL to identify callback port 1455 or 1457, and create the second tunnel from that local port to the actual Web port. Alternatively, cancel that Web operation and start a new one with the CLI; the CLI output belongs to the new operation and must not be used for the existing Web operation. Quota errors and catalog failures are reported as-is and never fall back to a paid provider. This compatibility path is experimental: the upstream interface may change.

Authentication is off by default — Cognivoy runs single-user. Turn it on and one data/ tree hosts an admin workspace, isolated per-user workspaces, and partner workspaces side by side:

data/
├── user/                    # Admin workspace + global settings
├── users/<uid>/             # Per-user scope: chat history, memory, notebooks, KBs
├── partners/<id>/workspace/ # Partner (synthetic-user) scope
├── cli-apps/                # Installed CLI apps, mounted read-only into the sandbox
└── system/                  # auth · grants · audit · user-secrets/<owner> (OAuth tokens)

The first registered user becomes admin and owns model catalogs, provider credentials, shared knowledge bases, skills, and per-user grants. Everyone else gets an isolated workspace and a redacted Settings page — admin-assigned models, KBs, and skills show up as scoped, read-only options, never as raw API keys.

Enable it: turn auth on in data/user/settings/auth.json, restart cognivoy start, register the first admin at /register, then add users from /admin/users and assign models, KBs, skills, partners, tool/MCP/CLI-app policy, and code-execution access through grants.

PocketBase stays a single-user integration — keep integrations.pocketbase_url blank for multi-user deployments unless you've wired up an external user store.

cognivoy chat opens an interactive REPL; cognivoy run <capability> "<message>" fires a single turn and exits. Both speak the same --capability, --tool, --kb, and --config flags.

cognivoy chat                                              # interactive REPL
cognivoy chat --capability deep_solve --kb my-kb --tool rag
cognivoy run chat "Explain the Fourier transform" --tool rag --kb textbook
cognivoy run deep_research "Survey 2026 papers on RAG" \
  --config mode=report --config depth=standard

Everything the Web app does is here too — knowledge bases (kb), sessions (session), partners (partner), skills (skill), notebooks, memory, and config. Full list below.

Cognivoy is built to be operated by another agent. Add --format json to any run and each turn streams NDJSON — one event per line (content, tool_call, tool_result, done, …), every line tagged with its session_id. Runs are headless-safe: an ask_user pause with no TTY auto-resolves with an empty reply instead of hanging.

# One shot, machine-readable
cognivoy run deep_solve "Find d/dx[sin(x^2)]" --tool reason --format json

# Chain turns in one stateful session — capture the id, reuse it
SID=$(cognivoy run deep_research "Survey 2026 papers on RAG" \
  --config mode=report --config depth=standard --format json \
  | jq -r 'select(.type=="done").session_id')
cognivoy run deep_question "Quiz me on that survey" --session "$SID" --format json

The CLI-only package lives in packaging/cognivoy-cli. In this checkout, install it from source:

python -m pip install -e ./packaging/cognivoy-cli

It isn't published to PyPI yet, so the main Get Started section keeps the source-install path.

EduHub is an upstream community hub for sharing teaching-oriented agent skills — Socratic tutors, flashcard builders, essay feedback, exam blueprints, concept explainers, and more. Cognivoy includes a compatible client, so a bare slug or an eduhub: prefix resolves to it without additional configuration.

Find and install — in the browser, open Learning Space → Skills → Import from EduHub to browse the catalog and download a skill straight into your library. From the terminal:

cognivoy skill search "socratic tutor"               # search EduHub (the default hub)
cognivoy skill install socratic-tutor                # fetch → verify → register
cognivoy skill install eduhub:socratic-tutor@1.2.0   # pin a hub and a version
cognivoy skill list                                  # local skills with their hub provenance

Publish your own — package a SKILL.md and share it back to the community:

cognivoy skill login                                 # browser sign-in to EduHub
cognivoy skill publish ./my-skill                    # interactive: pick a track + tags, then upload
cognivoy skill update                                # roll back or release a new version

EduHub is also a standalone, ClawHub-compatible registry, so agents that aren't Cognivoy (Claude Code, Codex, …) can use it directly through the eduhub CLI — npx eduhub install socratic-tutor.

Whatever the source, every import passes the same safety gate before anything touches your workspace:

• the registry's security verdict is checked first — flagged packages are refused unless you pass --allow-unverified;
• archives are extracted defensively (zip-slip / zip-bomb guards) behind a text/script suffix whitelist, so binaries never land in the workspace;
• frontmatter is normalized to Cognivoy's schema and always: is stripped, so a downloaded skill can never force itself into every system prompt;
• provenance — hub, version, verdict, and install time — is written to .hub-lock.json for audits and updates.

In multi-user deployments, installing is admin-only: a new skill lands in the admin catalog and stays invisible to other users until a grant assigns it, so an admin can vet it before rolling it out.

Because Cognivoy speaks the open Agent-Skills format, ClawHub works as a first-class source too — it's built in alongside EduHub. Pick it with the hub prefix:

cognivoy skill search "git release notes" --hub clawhub
cognivoy skill install clawhub:git-release-notes@1.0.1

Add more registries in settings/skill_hubs.json: a type: "clawhub" entry points at any compatible HTTP API (EduHub and ClawHub both speak it), type: "command" wraps whatever fetch CLI a registry ships, and "default" chooses the hub used for bare slugs. All of them feed the same import gate.