---

name: everos description: "Use this skill for EverOS local-first Markdown memory service setup, HTTP memory and knowledge APIs, cascade/OME operations, and observability integration." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

EverOS Repo Skill

EverOS is a Python 3.12+ local-first memory runtime for AI agents and user chats. It stores source-of-truth memory as Markdown, uses SQLite for system state and queues, and projects searchable rows into LanceDB for BM25/vector/scalar retrieval.

Use this skill when the task involves installing EverOS, configuring a memory root, running the server, calling the memory or knowledge HTTP APIs, operating cascade/OME maintenance, or enabling tracing/metrics.

Quick route map

User intent	Read next
Install EverOS, create config files, inspect health, start the server, or run the demo	setup-and-config
Add/flush/search/get user or agent memory through /api/v2/memory/*, work with prompt slots or multimodal message content	memory-api
Upload, replace, patch, delete, list, or search knowledge documents and topics	knowledge-base
Understand Markdown/SQLite/LanceDB layout, run everos cascade, rebuild indexes, configure OME, trigger reflection	cascade-and-evolution
Configure structured logs, Prometheus /metrics, OpenTelemetry, Langfuse tracing, content-capture privacy, or request-id correlation	observability

Minimal public setup

python -m pip install everos
# optional extras:
python -m pip install 'everos[multimodal]'
python -m pip install 'everos[otel]'

everos init --root ~/.everos
$EDITOR ~/.everos/everos.toml
everos server start --root ~/.everos
curl http://127.0.0.1:8000/health

EverOS has no in-process public client mode for ordinary app use: start the service and call the HTTP API. The everos console script is the primary local operator surface.

Configuration capabilities

Base install supports CLI, config generation, FastAPI app construction, Markdown/SQLite/LanceDB runtime, keyword-oriented inspection, and deterministic demo preview. Real memory extraction requires an OpenAI-compatible LLM configured in [llm]. Vector/hybrid/agentic search, knowledge search, clustering, and some OME strategies require embedding and/or rerank providers. Multimodal content requires the multimodal extra plus [multimodal] provider settings; Office document parsing also needs LibreOffice on the server host. OpenTelemetry export requires the otel extra and [observability] settings.

Shared references

• Repository provenance records the source snapshot, package version, and evidence paths. Read it before deciding whether this skill is stale for a checkout.
• Package overview summarizes architecture, runtime surfaces, storage model, and capability tiers.
• Cross-cutting troubleshooting covers install/import, provider gates, config/root confusion, security, and eventual consistency issues that span sub-skills.

Safe helpers

Prefer sub-skill-owned scripts for focused tasks. The helpers are safe by default: they only inspect local imports, print schema/config information, or call a user-supplied running server. Any helper that writes memory or knowledge requires an explicit flag.

Important operating boundaries

• New integrations should use /api/v2; /api/v1 is a legacy alias.
• EverOS binds to 127.0.0.1:8000 by default and ships no built-in authentication. Put an authenticated gateway in front before exposing it beyond loopback.
• Markdown is the source of truth. SQLite and LanceDB are derived/rebuildable, but deleting all of .index/ also deletes buffered messages not yet extracted.
• /add and /flush write Markdown synchronously when extraction occurs, but LanceDB search is eventually consistent. Use retry/backoff or everos cascade sync when a deterministic local drain is needed.
• Do not claim live LLM, embedding, rerank, multimodal, or Langfuse behavior is verified unless the target environment has credentials and the relevant optional dependencies configured.