Cognee (cognee)

Memory of the AI agents as a knowledge graph: Cognee ingests documents and conversations (add), extracts entities and relations with the LLM into a graph (cognify) and answers semantic, graph and GraphRAG searches over it. Each dataset gets its own graph file and vector schema; users, tenants and dataset permissions isolate the spaces. LLM and embedding calls go through the LiteLLM gateway of the platform.

Generated from catalog/services/cognee.yaml: do not edit by hand.

Summary

Version	1.6.6
Use	install and configure, configure and connect
Layer	functional (AI catalog)
Category	AI memory
Subcategory	Knowledge graph and GraphRAG
Installation methods	python, podman
External connection	no
Sandbox only	no

Licence

SPDX identifier	Apache-2.0
Licence policy	core licence (MIT or Apache-2.0)
Licence exception	-
Licence note	Cognee 1.6.6 (tag v1.6.6, commit df2b4facd7c1, wheel cognee-1.6.6-py3-none-any.whl sha256 bee551bd81c8f619acfc345088a4601b694ff941b80feaee52e4e03a25834408), the UI (cognee-frontend) and the MCP server (cognee-mcp 0.5.7) of the same repository under its Apache-2.0 LICENSE; LadybugDB 0.19.0 (embedded graph engine, fork of Kuzu) is MIT. Every Python dependency of the release is under an OSI licence (MIT, BSD, Apache-2.0, ISC, MPL-2.0, PSF-2.0, Unlicense, LGPL-3.0 with exceptions for psycopg2). Cognee Cloud, the GLiNER demo extractor, Neo4j, Neptune and the cloud recipes of distributed/deploy are neither installed nor configured.
Platform component	cognee

Packages and artifacts

A package the project of the service publishes (its image, its binary archive, its Python or OS package) carries the licence of the service; a third-party package carries the licence written on its row, read from the official upstream metadata.

Kind	Package	Version	Licence	Details
container image	docker.io/cognee/cognee	1.6.6	Apache-2.0	image key cognee, used by api, digest sha256:e3db1d4d2f4b8f1fa3cc74c7d8e01151f60ac0aa9e63957d277068c89e25970e
container image	docker.io/cognee/cognee-ui	1.6.6	Apache-2.0	image key cognee-ui, used by ui, digest sha256:dd569dab8646e5ee6568f2fa9ea4e928937594beda57f316bc243e4514e09e8d
container image	docker.io/cognee/cognee-mcp	main-df2b4fa	Apache-2.0	image key cognee-mcp, used by mcp, digest sha256:7b308820d37b21d8f64b2ec3db2fec6a12aee13c92e442b7ee3b240e027d7bdc
Python package	cognee[api,postgres-binary,tracing]	1.6.6	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	ladybug	0.19.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	gunicorn	23.0.0	MIT	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	uvicorn	0.53.0	BSD-3-Clause	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	fastapi	0.141.1	MIT	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	fastapi-users	15.0.5	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	litellm	1.105.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	openai	2.54.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	sqlalchemy	2.0.54	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	alembic	1.20.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	asyncpg	0.31.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	psycopg2-binary	2.9.13	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	pgvector	0.3.6	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	opentelemetry-sdk	1.44.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	opentelemetry-exporter-otlp-proto-http	1.44.0	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	langdetect	1.0.9	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13
Python package	pixeltable-pgserver	0.7.1	Apache-2.0	interpreters, in order of preference python3.12, python3.11, python3.13

Images are pinned by tag in the descriptor, and by digest when the descriptor declares one (the offline bundle then refuses a pulled image that does not carry it); the digest is recorded in the mirror of the Manager when the offline bundle is imported, and every pull or load is checked against it.

Supported systems and architectures

Method	Operating systems	Architectures
python	RHEL family (dnf), Debian family (apt), Arch Linux (pacman)	x86_64, aarch64 (not restricted by the descriptor)
podman	RHEL family (dnf), Debian family (apt), Arch Linux (pacman)	x86_64, aarch64 (not restricted by the descriptor)

Certified host systems: RHEL 8/9/10, Rocky Linux 8/9/10, AlmaLinux 8/9/10, Oracle Linux 8/9, CentOS Stream 9/10, Ubuntu 22.04/24.04, Debian 12, Arch Linux (rolling). The host preflight accepts these systems (other releases of a known family give a warning, end of life releases are refused) and the x86_64 and aarch64 architectures.

Prerequisites

Host

• at least 4096 MB of memory (blocking, memory).
• at least 8192 MB of memory (warning, memory_comfort).
• at least 2 CPU (blocking, cpu).
• at least 10240 MB free for /var/lib/navin (blocking, disk).

Network and ports

Component	Port name	Port	Opened in the host firewall
api	api	8420/tcp	yes
api	api_local	8420/tcp	no
ui	ui	8421/tcp	yes
mcp	mcp	8422/tcp	no

Services and connections needed

• database: postgresql.
• llm: litellm.
• otel: otel-collector (optional).

Details and resolution: see Dependencies.

Installation methods and modes

Method	Default	What the method uses
python	yes	a Python virtual environment holding the Python packages of the descriptor
podman	-	the container images of the descriptor, run by Podman under systemd

Components

Component	Kind	Installation methods	Ports
Cognee API (api)	master	python, podman	8420/tcp, 8420/tcp
Cognee UI (ui)	master	podman	8421/tcp
Cognee MCP server (mcp)	master	podman	8422/tcp

Single server (single) (default)

One API (one gunicorn worker) owning the LadybugDB files of the host, with the optional UI and MCP server next to it.

What the mode deploys:

Component	Placement
api	min 1, max 1
ui	min 1, max 1
mcp	min 1, max 1

Install this service

The install wizard of the Navin Manager plans, validates and runs the installation; you need the services:install permission on the cluster.

1. Open the wizard. In Clusters, choose the row action Install services of the target cluster. The wizard opens on its first step, Service.

2. Choose the service. On the Service step, select Cognee. The card shows the licence Apache-2.0 and the dependencies the installation needs. Pick the Method among python, podman (default python).

3. Choose the hosts and the mode. On the Hosts step, tick the hosts, then choose the Mode: single (default). Each mode deploys the components listed in Installation methods and modes. With several components, place each one on its hosts in Component placement (Suggest placement proposes a placement that respects the cardinalities).

4. Bind the dependencies. On the Bindings step, choose for each dependency (database, llm, otel) Install with the platform or Use an existing connection; an optional dependency can stay on No binding.

5. Configure. On the Configure step, review the parameters (see Configuration parameters). Passwords left empty with a generated default are generated and sealed by the Manager.

6. Review and install. On the Install step the Manager validates the plan and shows the installation order; when it says "The plan is valid. Ready to install.", choose Install. A highly available mode is installed only when its HA checks pass.

7. Verify. The operation ends when the health checks of every instance pass (see Health checks); the service then shows as running in Services. Open the service and choose Run the smoke test in the Checks card: In a dataset of its own (navin-smoke-<nonce>): checks that a call without key and a public registration are refused, adds two texts, runs cognify (LLM graph extraction and embeddings), finds a chunk by semantic search, asks a graph completion, reads the graph of the dataset, then deletes the dataset whatever happened.

Connect an existing service

Schemes	https, http
Default port	8000
Authentication	required
Health check	/health

Authentication method	Fields	Description
api_key	api_key	API key of a Cognee user (POST /api/v1/auth/api-keys), sent as X-Api-Key; the server must run with REQUIRE_AUTHENTICATION and ENABLE_BACKEND_ACCESS_CONTROL true.

Official documentation: https://docs.cognee.ai/http_api

High availability

Cluster	none (none)
High availability	none (none)
Failover	none (none)
Replication	none (none)
Quorum	none (none)
Discovery	none (none)
Entry point	any node (any-node)
Official documentation	https://docs.cognee.ai/setup-configuration/graph-stores
References	https://github.com/topoteretes/cognee/blob/v1.6.6/entrypoint.sh, https://github.com/topoteretes/cognee/blob/v1.6.6/cognee/infrastructure/databases/graph/postgres_demo/adapter.py, https://github.com/topoteretes/cognee/blob/v1.6.6/cognee/infrastructure/databases/dataset_database_handler/supported_dataset_database_handlers.py

Topology per mode

Modes	Highest catalog level	Distribution	Quorum component	Fault domains	Description
single	standalone (standalone)	single (single)	-	1	One API owning the embedded LadybugDB graph files: its loss stops the memory API until it is restarted or restored; the relational metadata and the vectors stay in PostgreSQL.

Notes

• Standard profile (PostgreSQL with pgvector and embedded LadybugDB): single instance only. LadybugDB is an embedded engine whose files are locked by one process (the official entrypoint runs one gunicorn worker); two APIs would each hold a different graph. No distributed or replicated mode is offered.
• External graph backends are not offered: Neo4j, Neo4j Aura and Neptune are excluded (licence or cloud service), and the PostgreSQL graph backend of Cognee 1.6.6 (postgres_demo) is a demo the source code calls not production ready. Turso is not a validated backend of the platform.
• With the highly available PostgreSQL service, a lost host is replaced by a reinstall and the restore of the last backup (dump, LadybugDB files and data root taken together).
• The API exposes no metrics endpoint (no Prometheus route in 1.6.6): the observability uses blackbox probes of /health and the authenticated service check; traces go to OpenTelemetry when bound.

Configuration parameters

Network

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
API port (port)	port	8420	min 1024	no	no	yes	Port of the REST API (8000 upstream, moved off the ports of vLLM and of the other services).
API exposure (listen_scope)	enum	network	network, local	no	no	yes	network: the API listens on the private address of the host only (never on every address) and the host firewall opens it to the consumers; local: loopback only, for consumers on the same host. It is never published by a gateway.
TLS (tls_enabled)	bool	false	-	no	no	yes	Serve the API over TLS (gunicorn --certfile and --keyfile) with the certificate the Manager issues to the host; the binding of the consumers then uses https and the authentication cookie is secure only.
UI public address (public_url)	string	(empty)	^(|https?://[A-Za-z0-9.-]+(:[0-9]{1,5})?)$	no	no	yes	Address the users open for the UI behind the gateway, for example https://cognee.bank.internal: it is the only origin allowed by the CORS rules of the API (credentials included) and, in https, the authentication cookie is secure only. Empty: the address of the UI port of the host.
Request timeout (request_timeout_s)	int	3600	min 60, max 86400, seconds	no	no	yes	Seconds a request may run before gunicorn restarts the worker (--timeout): a synchronous cognify of a large dataset is long.

Models

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
LLM (llm_model)	string	qwen-local	^[A-Za-z0-9._:/@-]{1,128}$	no	no	yes	Model name of the gateway used for the graph extraction (JSON structured output), the summaries and the answers; default qwen-local, the default served model name of a vLLM engine bound to LiteLLM (the gateway has no default entry, audit P5B-010).
LLM answer tokens (llm_max_tokens)	int	4096	min 256, max 131072	no	no	yes	Largest completion of one LLM call (LLM_MAX_COMPLETION_TOKENS, 16384 upstream): keep it within the context of the local model.
Embedding model (embedding_model)	string	bge-m3	^[A-Za-z0-9._:/@-]{1,128}$	no	no	yes	Embedding model name of the gateway (OpenAI compatible /v1/embeddings, no dimensions parameter sent); default bge-m3, served by a vLLM engine in embedding mode bound to LiteLLM.
Vector size (embedding_dimensions)	int	1024	min 64, max 16000	no	no	yes	Size of the vectors the embedding model returns (1024 for bge-m3): the vector columns of the datasets are created with it. Changing it needs new datasets (or a restore into an empty installation).
Embedding input tokens (embedding_max_tokens)	int	8192	min 128, max 131072	no	no	yes	Tokens the embedding model accepts per input (EMBEDDING_MAX_COMPLETION_TOKENS, 8192 for bge-m3): the chunks are limited to it. Token counts use the bundled tiktoken encodings: no tokenizer is downloaded.
Gateway key (llm_api_key)	password	(empty)	-	no	yes	no	LiteLLM virtual key of Cognee (sk-...), the way to meter and cap its spend; empty, the master key of the bound gateway is used.
LLM rate limit (llm_requests_per_minute)	int	0	min 0, max 100000	no	no	yes	LLM and embedding calls per minute of the whole installation (LLM_RATE_LIMIT_* and EMBEDDING_RATE_LIMIT_*); 0: no limit of Cognee, the gateway still applies the limits of its key.

Processing and limits

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
Chunk size (chunk_size)	int	1500	min 100, max 32000	no	no	yes	Default size of the chunks cognify cuts the documents into (CHUNK_SIZE, in tokens; a request may give its own chunk_size).
Chunk overlap (chunk_overlap)	int	10	min 0, max 2000	no	no	yes	Tokens repeated between two consecutive chunks (CHUNK_OVERLAP).
Concurrent pipelines (max_concurrent_pipelines)	int	2	min 1, max 64	no	no	yes	Datasets processed at once by the dataset queue (DATASET_QUEUE_MAX_CONCURRENT); the others wait. cognify is CPU and memory heavy (parsing, chunking, graph merge in the embedded LadybugDB worker): raise it only with the memory to match. (recommended clamp(floor(host.cpu_count / 4), 1, 16))
Graph buffer pool (graph_buffer_pool_mb)	int	512	min 64, max 65536, MB	no	no	yes	Buffer pool of each open LadybugDB graph (KUZU_BUFFER_POOL_SIZE): the memory reserved per dataset graph being read or written.
Automatic feedback (auto_feedback)	bool	false	-	no	no	yes	Analyse every answered search of a session with one more structured LLM call to keep guidance in the session (AUTO_FEEDBACK, on upstream); off by default to halve the LLM cost of the searches.
API keys per user (max_api_keys_per_user)	int	10	min 1, max 100	no	no	yes	Quota of API keys a user may create (MAX_USER_API_KEYS).

Storage and retention

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
Session retention (session_ttl_days)	int	7	min 1, max 3650, days	no	no	yes	Days the question and answer history of a search session is kept in the session cache (SESSION_TTL_SECONDS); the datasets, graphs and vectors are kept until they are deleted through the API.

UI and MCP

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
Web UI (ui_enabled)	bool	false	-	no	no	yes	Run the official UI (cognee-ui, Next.js, method podman only) next to the API: login with the account of the administrator, datasets, graph explorer and searches. The browser calls the API directly, so the API must be reachable from the users (listen_scope network).
UI port (ui_port)	port	8421	min 1024	no	no	yes	Port of the UI (plain HTTP: publish it through the TLS gateway of the platform).
MCP server (mcp_enabled)	bool	false	-	no	no	yes	Run the official MCP server (cognee-mcp, method podman only) in API mode with the service API key: tools remember, recall, search and cognify for the agents of the same host. It has no authentication of its own, so it listens on the loopback of the host only.
MCP port (mcp_port)	port	8422	min 1024	no	no	yes	Port of the streamable HTTP endpoint /mcp of the MCP server, 127.0.0.1 of the host only.

Security

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
Administrator e-mail (admin_email)	string	cognee-admin@navin.internal	^[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}$	no	no	no	Login of the default user of Cognee (DEFAULT_USER_EMAIL, superuser, owner of the service API key); the default_user@example.com account of the upstream code is never created.
Administrator password (admin_password)	password	generated	-	no	yes	no	Password of the default user (DEFAULT_USER_PASSWORD), generated at each install: Cognee applies it once, to an account without password; afterwards it is changed in Cognee only (rotation by the administrator).
Service API key (service_api_key)	password	generated	-	no	yes	no	API key of the default user installed by the launcher at each start (X-Api-Key header, stored hashed), given to the consumers as the api_key secret of the api endpoint and used by the MCP server and the smoke test; rotated by a new value and a restart.
JWT secret (jwt_secret)	password	generated	-	no	yes	no	FASTAPI_USERS_JWT_SECRET signing the login tokens and cookies; changing it ends every session.
Password reset secret (reset_password_secret)	password	generated	-	no	yes	no	FASTAPI_USERS_RESET_PASSWORD_TOKEN_SECRET of the password reset tokens.
Verification secret (verification_secret)	password	generated	-	no	yes	no	FASTAPI_USERS_VERIFICATION_TOKEN_SECRET of the e-mail verification tokens.
API key salt (api_key_salt)	password	generated	-	no	yes	no	API_KEY_HASH_SALT of the stored API keys (HASH_API_KEY on); changing it invalidates every API key but the service one, which the launcher reinstalls.
Login lifetime (token_lifetime_minutes)	int	60	min 5, max 1440, minutes	no	no	yes	Lifetime of a login token or cookie (JWT_LIFETIME_SECONDS).
Open registration (allow_registration)	bool	false	-	no	no	yes	Let anyone who reaches the API create an account (POST /api/v1/auth/register, open upstream). Off: the launcher answers 403 unless the request carries the service API key, so the accounts are created by the administrator or by Navin Enterprise.
Ingestion by URL (allow_url_ingestion)	bool	false	-	no	no	yes	Let add fetch web pages and clone repositories given as URLs (ALLOW_HTTP_REQUESTS; private addresses stay refused by the SSRF protection of Cognee). Off: only uploaded files and texts.
Cypher queries (allow_cypher)	bool	false	-	no	no	yes	Allow the search type CYPHER (raw graph queries, ALLOW_CYPHER_QUERY, off upstream).

Observability

Parameter	Type	Default	Allowed values	Required	Secret	Restart	Description
Log level (log_level)	enum	WARNING	ERROR, WARNING, INFO, DEBUG	no	no	no	Verbosity of Cognee and of gunicorn (LOG_LEVEL).
Tracing (tracing_enabled)	bool	false	-	no	no	yes	Export the OpenTelemetry spans of Cognee (COGNEE_TRACING_ENABLED) to the bound OpenTelemetry collector (OTLP over HTTP, python method only).

Dependencies

For each active dependency the Manager takes the first of: the binding chosen in the wizard; a service of the same installation that provides it; a service already installed in the cluster; the only connection of the expected type. Otherwise an optional dependency stays unbound, and a required one is added to the installation in its default mode (the column below names the catalog service added).

Binding	Provided by	Connection type	Endpoint	Optional	When nothing is bound	Description
database	postgresql (versions >= 14)	postgresql	sql	no	postgresql is added to the installation	Dedicated database cognee: users, datasets, permissions and pipeline state (alembic migrations) and the pgvector collections of every dataset in its own schema ds_<dataset> of the same database. The provisioning creates the role cognee, owner of this database only (no superuser, no CREATEDB), and the vector extension.
llm	litellm	-	api	no	litellm is added to the installation	OpenAI compatible gateway of the LLM (graph extraction, summaries, answers) and of the embedding model; the vLLM and Ollama engines of the Manager are reached through it. The launcher refuses to start when the gateway does not list both models.
otel	otel-collector	-	otlp_http	yes	stays unbound (optional)	OpenTelemetry traces of the pipelines and searches (OTLP over HTTP, python method only: the official image has no OpenTelemetry SDK).

Checks of the bound endpoints

• the endpoint /v1/models of the binding llm answers with the status 200 and lists the values of llm_model, embedding_model, checked by the Manager before the installation (blocking, llm_models).

Endpoints

Endpoint	Scheme	Component	Port	Path	TLS parameter	Binding parameters	Binding secrets	Provisioning command
api	http	api	8420 (port)	/api/v1	tls_enabled	admin_email, graph_backend, vector_backend, access_control, embedding_dimensions	api_key	-

Client configuration files

Files rendered for the client configuration download:

• cognee.env (client.env.j2)

Commands

Command	Component	Target	Description
Detailed health (health_detailed)	api	first host	Health of every store and of the model gateway as the API sees it (GET /health/detailed: relational, vector, graph, file storage, LLM and embedding).
Apply the migrations (migrate)	api	first host	Applies the relational migrations (alembic) and the graph and vector revisions of the release (cognee-cli upgrade, also done at every start of the API). Method python; with podman the API applies them at its start.
Recent logs (logs)	api	chosen host	Last lines of the API journal.
UI logs (ui_logs)	ui	chosen host	Last lines of the UI journal.
MCP logs (mcp_logs)	mcp	chosen host	Last lines of the MCP server journal.

Health checks

Liveness (per instance)

Component	Check	Details
api	HTTP {{ 'https' if config.tls_enabled else 'http' }}://{{ host.fqdn if config.tls_enabled and config.listen_scope == 'network' else ((host.ip or '127.0.0.1') if config.listen_scope == 'network' else '127.0.0.1') }}:{{ config.port }}/health	Expected 200; Expected "status":"ready"; 120 retries; interval 3 s; timeout 10 s
ui	HTTP http://{{ host.ip or '127.0.0.1' }}:{{ config.ui_port }}/api/runtime-config	Expected 200; 60 retries; interval 2 s; timeout 5 s
mcp	TCP 127.0.0.1:{{ config.mcp_port }}	60 retries; interval 2 s; timeout 5 s

Readiness (service check)

Check

Details

{{ (paths.opt ~ '/venv/bin/python') if service.method == 'python' else '/usr/bin/python3' }} {{ paths.opt }}/app/navin_cognee.py check {{ paths.etc }}/api.env {{ 'https' if config.tls_enabled else 'http' }}://{{ host.fqdn if config.tls_enabled and config.listen_scope == 'network' else ((host.ip or '127.0.0.1') if config.listen_scope == 'network' else '127.0.0.1') }}:{{ config.port }}/api/v1/datasets {{ tls.ca_file if config.tls_enabled else '' }}

10 retries; interval 3 s; timeout 15 s

Smoke test

In a dataset of its own (navin-smoke-<nonce>): checks that a call without key and a public registration are refused, adds two texts, runs cognify (LLM graph extraction and embeddings), finds a chunk by semantic search, asks a graph completion, reads the graph of the dataset, then deletes the dataset whatever happened. (api: refused_without_key, no_anonymous_account, add, cognify, semantic_search, graph_completion, graph, cleanup)

Preflight checks

Check	Type	Meaning	Severity	Components
memory	memory_mb {"min": 4096}	at least 4096 MB of memory	blocking	api
memory_comfort	memory_mb {"min": 8192}	at least 8192 MB of memory	warning	api
cpu	cpu_count {"min": 2}	at least 2 CPU	blocking	api
disk	disk_free_mb {"min": 10240}	at least 10240 MB free for /var/lib/navin	blocking	api
api_port_free	port_free {"param": "port"}	port 8420 free on the host	blocking	api
ui_port_free	port_free {"param": "ui_port"}	port 8421 free on the host, only when ui_enabled is true	blocking	ui
mcp_port_free	port_free {"param": "mcp_port"}	port 8422 free on the host, only when mcp_enabled is true	blocking	mcp
llm_models	http_reachable {"binding": "llm", "path": "/v1/models", "bearer_secret": "token", "contains_params": ["llm_model", "embedding_model"], "timeout_s": 10}	the endpoint /v1/models of the binding llm answers with the status 200 and lists the values of llm_model, embedding_model, checked by the Manager before the installation	blocking	api

Backup and restore

The API is stopped first (its shutdown checkpoints the write ahead log of every LadybugDB file), then pg_dump (custom format, without owner nor privilege) of the dedicated database (relational metadata and the pgvector schema of every dataset) and tar archives of the system root (LadybugDB graph files, session cache) and of the data root (stored documents), with a manifest of their sha256; the API is started again. verify checks every sha256, reads the table of contents of the dump and every member of both archives. Restore, API stopped: in one transaction DROP OWNED BY the role of Cognee and the objects of the dump (the vector extension stays the one of the provisioning), then both roots are replaced.

Format	cognee-pgdump-ladybug
Component	api
Files	{{ backup.dir }}/cognee.dump, {{ backup.dir }}/system.tar.gz, {{ backup.dir }}/data.tar.gz, {{ backup.dir }}/manifest.json
Restore test	yes
Instances stopped for the restore	yes

Upgrade

Supported from
Stateful	yes
Backup before the upgrade	yes
Rollback	no (The migrations of Cognee (alembic and the revisions of the LadybugDB storage format) only go forward and the Manager never downgrades them: the way back is the restore of the backup taken before the upgrade.)

Uninstall

Paths removed when the data is purged: {{ paths.data }}, {{ paths.log }}.

Quick links

• API reference: {{ 'https' if config.tls_enabled else 'http' }}://{{ host.fqdn }}:{{ config.port }}/docs
• Cognee UI: {{ (config.public_url ~ '/dashboard') if config.public_url else ('http://' ~ host.fqdn ~ ':' ~ config.ui_port ~ '/dashboard') }}

Upstream documentation

• Homepage: https://docs.cognee.ai
• Source: https://github.com/topoteretes/cognee/tree/v1.6.6
• High availability: https://docs.cognee.ai/setup-configuration/graph-stores
• High availability: https://github.com/topoteretes/cognee/blob/v1.6.6/entrypoint.sh
• High availability: https://github.com/topoteretes/cognee/blob/v1.6.6/cognee/infrastructure/databases/graph/postgres_demo/adapter.py
• High availability: https://github.com/topoteretes/cognee/blob/v1.6.6/cognee/infrastructure/databases/dataset_database_handler/supported_dataset_database_handlers.py
• Authentication: https://docs.cognee.ai/http_api