Start in 60 seconds.
Go as deep as you want.
Three ways to use chanio, in order of commitment. Each one works on its own. Each one makes the next more powerful. You never have to climb the whole ladder — Tier 1 is enough to feel it.
See a sample first — 10 seconds, nothing to installThe Lab
Paste a stream of what you've been thinking, reading, or working on. chanio finds the question you've been circling but never asked — and fills it. Runs Claude in your browser using your own key; nothing leaves your machine except the call to Anthropic.
- 1Open the Lab — see a real sample first, no key required.
- 2Paste your Anthropic key (set a $5 cap on it; a run costs ~$0.01–0.05).
- 3Drop in your stream. Get a gap-fill, a 3-round Deep Dive, or patterns across sessions.
- 4Export your whole archive as a file you own, anytime.
The Extension
A quiet browser extension sits on top of Claude, ChatGPT, Gemini, and Perplexity. Every prompt you send is captured to a local archive — no copy-paste, no thinking about it. Your knowledge graph compounds on its own while you work.
- 1Install the extension (one zip, ~13 KB, Manifest V3).
- 2Use your AIs exactly as you already do.
- 3chanio captures only what you type into an AI — never your general browsing, passwords, or private tabs.
- 4Your archive grows passively, stored on your device.
chanio-core + Obsidian
The terminal layer. chanio-core captures every Claude Code session automatically, and a nightly job compiles everything into your Obsidian vault — a morning brief plus a living state note for each project you're working on. An MCP connector exposes the whole graph to any AI that speaks the protocol. This is "storage comes home" in full.
- 1Point chanio-core at your Obsidian vault.
- 2Every terminal AI session is captured to disk automatically.
- 3A nightly compile turns raw sessions into briefs + per-project state notes.
- 4Any AI reads your graph over MCP — you bring the context, not the platform.
~$0.50/day across 1,587 vault files.
Same principle that took a mid-tier model past the frontier on Frontier SWE in June 2026: the harness around the model is where the leverage is. Zenith, Intelligent Internet proved it for coding agents. Google's June 2026 whitepaper The New SDLC made it doctrine: Agent = Model + Harness — the model is the smallest part of the system, and the harness is the part you own. Their other finding: with implementation nearly free, the bottleneck moves to what humans specify and validate. chanio is the same bet at personal scale — your context is the specification. Own the surround, rent the intelligence.
Most personal-AI products cost $20–100/month because they rent intelligence on every request. chanio costs $0.50/day because it owns its knowledge and rents intelligence only at the synthesis step. The canonical pass — extraction, deduplication, entity resolution — runs on local hardware at $0. Cloud models handle only the steps where frontier-model quality actually changes the output.
| Call | Model | Where | Marginal cost |
|---|---|---|---|
| Per-session extract | ollama / ornith:35b | Local | $0runs on your hardware |
| Nightly canonical | ollama / gemma3:27b | Local | $0deduplication + entity merge |
| Morning brief | claude-opus-4 | Cloud | ~$0.50/daythe measured number |
| Chat (/api/chat) | claude-sonnet-4-5 | Cloud | per-messageonly when you ask |
| Backfill import | claude-sonnet-4-5 (BYOK) | Cloud — your key | $0 to chanioyour Claude.ai export, your bill |
Own knowledge, rent intelligence. Local models do the canonical pass — extraction, deduplication, embedding — because that is where ownership matters. Cloud models do synthesis — brief, chat, import — because that is where frontier-quality changes the output. No call crosses that boundary in the wrong direction.
Extracted facts never leave your machine for the canonical pass. The local model sees everything; the cloud model sees only the synthesis input.
The morning brief compiles at 5 AM without waiting on cloud queues. Local inference is bounded only by your hardware.
60x cheaper than alternatives at personal-vault scale. 1,587 files processed nightly at $0 marginal cost.
chanio survives your API-key rotation. BYOK means your keys, your bill, your data. The vault stays yours even if chanio disappears.
The 30-day backfill option turns your Claude.ai JSONL exports into tomorrow morning's dense brief — one import, zero recurring cost.
See the backfill optionOnly at Tier 3. The Lab and the extension store your data without it — the Lab in your browser (exportable to a file), the extension on your device. Obsidian is the upgrade for people who want their whole graph in a tool they fully control, browsable forever, with a nightly compile running on their own machine. Start without it. Add it when you want the full graph at home.
The fastest way to get it
is to feel it.
Open the Lab, paste a stream of what you've been thinking, and watch it find the question you didn't ask. No install. 60 seconds.
Open the Lab →Compute stays rented. Storage comes home.