FAQ

How much does it cost to run?

Self-hosted: ~$6-24/mo for a DigitalOcean droplet (the $6 nano works for small vaults; $24 4GB droplet recommended once you have a few thousand thoughts).

LLM API costs: ~$2-10/mo depending on vault size and synthesis cadence. Most workloads route through cheap models (DeepInfra Qwen) via llm_router. There are no subscription fees — you only pay your infra + LLM provider directly.

What LLMs does it use?

By default, DeepInfra (Qwen) for embeddings and synthesis (cheap and good enough for ~95% of tasks), and Anthropic Claude as fallback for complex reasoning (synthesis of contradictions, multi-perspective questions). The llm_router picks the cheapest capable model automatically based on the task type and falls back if a provider fails.

You can also point it at a local Ollama instance — everything will run on-device.

Is my data private?

Yes. Everything runs on your own server. Your notes, meetings, and graph never leave your infrastructure unless you configure a cloud LLM API (Anthropic, OpenAI, DeepInfra) — in which case the text you send for synthesis goes to that provider under their API terms.

If you want zero data egress, use a local LLM (Ollama) for synthesis. Performance trade-off applies.

You control the data. Exocortex itself doesn’t transmit anything outside your server.

If you use cloud LLM APIs: their API terms apply to the text you send. Anthropic and OpenAI both offer zero-retention options on their API tier — enable those.
If you use self-hosted LLMs (Ollama or DeepInfra private): everything stays local.
For EU users processing third-party personal data (e.g. meeting notes mentioning clients), the same rules apply to Exocortex as to any tool you self-host. You are the data controller.

How do I back up my data?

Two things to back up: PostgreSQL and the vault directory.

pg_dump exocortex > backup.sql
tar czf vault-$(date +%Y%m%d).tar.gz /opt/exocortex-vault

Schedule both daily, ship off-server. That’s it.

What happens if the server goes down?

Your Obsidian vault continues to work offline — it’s just .md files. Syncthing (or whatever sync layer you use) re-syncs when the server is back up. The Postgres graph rebuilds from the vault on next ingest if you ever lose the DB entirely.

No data loss as long as you have either the vault or the Postgres dump.

How is this different from a RAG chatbot?

RAG chatbots answer questions by retrieving chunks and stuffing them into an LLM prompt. They forget between sessions.

Exocortex builds structured reasoning over time:

Typed edges (supports, contradicts, derives_from, supersedes) — not just embedding similarity
Autonomous synthesis at night — clusters of related thoughts become Markdown perspectives
Contradiction detection — when two sources disagree, the graph surfaces it
Provenance tracking — human-written knowledge ranked higher than AI-generated to prevent feedback loops

It gets smarter the longer you use it, not just bigger.

Can I use it with Obsidian?

Yes. The vault is just a directory of .md files with frontmatter. Obsidian reads it directly, and Exocortex writes back to it. Wikilinks, tags, and Dataview queries all work.

Does it work without a Notion account?

Yes. Notion is optional (used by the Two-Way Cockpit feature for mirroring projects/tasks). Core functionality — synthesis, MCP, briefings, Gap Radar — works with just Obsidian + Telegram.

Is it open source?

Apache 2.0 + Commons Clause.

✅ Free for personal use
✅ Free for non-commercial community use
✅ Modify, fork, run, share
❌ Selling services or products built on Exocortex requires a commercial license

See the license explainer for plain-language details.

FAQ