Summary of "Моя жизнь разделилась на ДО и ПОСЛЕ: OpenClaw + Obsidian"

High-level summary

This session is a practical walkthrough/demo of using a local agent platform (referred to in the subtitles as OpenClow / OpenCl / “OpenClaw”) together with Obsidian as a personal “second brain.” The presenter (Rustam, introduced by Max) demonstrates how agents can read, write, and act on your Obsidian vault and other systems (APIs, browser, email, Telegram, file system). The talk shows real-world automation patterns, security trade-offs, and product/technical tips.

Core message: Digitize and structure your personal content (notes, transcripts, podcasts, drafts) into Obsidian first; then grant a well-configured, scoped agent access so it can automate summarization, tagging, generation, testing, deployments and other workflows. Agents amplify productivity but introduce real security and governance risks if given broad, unchecked access.

Key technological concepts and tools

Agent platforms
- OpenClaw/OpenCl (desktop Node.js agent that talks to LLMs and runs sub-agents)
- Antigravity (agent GUI / orchestration tool)
- “Wipe-coding” / wipecoding: agent-assisted rapid coding pattern
- General agent pattern: agents configured by markdown/prompt files
Large models / backends: Anthropic (Claude), Google Gemini, OpenAI ChatGPT, local LLMs — the agent orchestrates between models and systems
Obsidian: canonical personal vault (Markdown files). Agent reads the vault, creates notes, tags AI-generated content, and preserves provenance
Voice / speech: TTS and STT integrations (Opus 4.6, Aquavoice, Supervoice) for phone/voice interfaces
Third-party connectors/APIs: Parallel AI (research), Banana.dev (image generation), others via API keys; agents can call APIs and write glue scripts
Developer tooling: automated browser testing (Chrome DevTools), running scripts/commands, generating bug reports, making small deployments, scripted database migrations
Files-first philosophy: keep data in Markdown/txt (no proprietary lock-in), maintain backups (e.g., Synology), map agent threads to vault folders to avoid context mixing

Practical features, use-cases and workflows

Transcription → Structured notes
- Upload meeting transcript; agent summarizes and writes: main summary, five key ideas, suggested follow-ups; stores them in the relevant client/project folder in Obsidian.
Personal second brain + agent collaborator
- Agent scans the vault to build a semantic profile of “you,” proposes content ideas, writes drafts into Obsidian, and tags AI-generated notes.
Multi-thread / scoped contexts
- Create separate agent threads for each project (mapped to specific Obsidian folders) so context doesn’t bleed between clients, blog drafts, codebases, etc.
Generative media pipeline
- Request a batch of images in a chosen style → agent calls image API (Banana/Gemini), stores assets in the appropriate folder, can generate English and local-language variants.
Code QA / web app testing
- Agent opens the app, runs browser-based tests via Chrome DevTools, finds bugs, writes repair instructions or pull-request-ready docs, and can (optionally) run fix scripts.
App automation and infra tasks
- Example: modify DB to add a timezone field — agent inspects the schema, generates migration scripts, and can run them (with appropriate access).
Content repurposing & deployment
- Convert repeatable advice from one-on-ones into a checklist → HTML → deploy to a website using the agent.
On-the-go ideation
- Voice note while walking → agent researches current web resources (Parallel AI), drafts concept sketches, and saves them to Obsidian.
Phone / voice interface
- Make a phone call to the agent; it uses STT → LLM → TTS to respond (e.g., a short pep talk or summary).

Security, governance, and setup guidance

Danger summary
- If given broad permissions, agents can read/send email, access passwords/crypto keys, purchase products, and exfiltrate data. (Examples included an agent buying an expensive course overnight.)
Mitigations and operational tips
- Give minimal, scoped permissions only.
- Isolate agent networking and ports.
- Avoid exposing sensitive folders or accounts; do not store critical secrets unencrypted on the machine.
- Restrict Telegram/email sources to trusted senders only.
- Tag agent-created content (e.g., add a tag marking AI-generated files) to preserve provenance.
- Use separate threads/contexts per project and map each to specific Obsidian folders.
- Keep backups (weekly snapshots to an offline NAS / Synology).
- Expect fast-moving change in tools—re-evaluate configurations and behavior frequently.

Tool comparisons and practical notes

OpenClaw/OpenCl
- Flexible local orchestrator integrating tightly with Obsidian and external APIs.
- Good for continuous background work and multi-step subagent workflows.
Antigravity
- Praised for hands-on coding workflows and deep coding help.
- Opus 4.6 audio model was noted in relation to coding/audio tasks.
Wipe-coding (wipecoding)
- Pattern where agents write and iterate code and technical specs rapidly; useful for prototyping and releasing projects quickly.
Audio/voice notes
- Opus 4.6: powerful audio model but compute- and token-heavy—best for surgical audio tasks.
- STT options: Aquavoice / Supervoice highlighted for quality.

Product / tutorial takeaways (how-to checklist)

Before installing an agent

Digitize and structure your content (Obsidian vault, Markdown).
Decide rules for what content the agent may author or summarize.

Basic setup steps showcased

Install the desktop agent and select an LLM backend (cloud or local).
Create project-specific threads and map each to an Obsidian folder.
Provide API keys for services the agent should use (image generation, Parallel AI).
Configure STT/TTS tokens for voice features if needed.
Test with non-sensitive tasks (generate images, summarize a meeting) before expanding access.
Tag and mark any AI-written files; keep provenance and versioning.

Two essential questions to ask the agent

“How to do what I need?” — request step-by-step instructions or scripts.
“Explain how this works” — force the agent to teach the logic/architecture so you don’t blindly delegate.

Warnings and human factors

Eliza effect reminder: people may unconsciously treat agents as understanding/human; retain skepticism and human-to-human communication where necessary.
Cognitive load and burnout: agents reduce friction and context switches, but can create overflow (more tasks, less rest). Set boundaries.
Competence vs. delegation: agents amplify capability but do not replace core understanding. Use agents to automate after you learn the underlying logic; ask them to explain implementations.

Concrete demos mentioned

Live phone call to an agent producing a short spoken motivational message.
Agent created a LinkedIn post image set in a specific style and saved it to Obsidian.
Agent scanned Obsidian, proposed unpublished post ideas, and saved them with tags.
Agent reviewed a web deployment, found bugs, prepared edit docs for developers, and added them to project files.
Agent produced a client meeting summary, “what could be improved” notes, and an agenda for the next meeting; saved everything in the correct vault folder.
Agent scripted database updates (e.g., add a timezone field across records) and populated fields (demo finished in ~5 minutes).

Philosophy & final recommendations

Treat your digital footprint (notes, talks, transcripts, posts) as capital: digitize it and store it in structured Markdown first.
Use agents to reduce friction in repetitive tasks, but define clear rules about what to human-verify versus what to delegate.
Maintain data control and backups; prefer files-first storage and avoid proprietary lock-in.
Stay agile: tools evolve quickly—revisit earlier experiments regularly.

Main speakers / sources

Rustam — primary presenter (demonstrations, technical walkthroughs, examples, tooling opinions).
Max — introducer and moderator (context, product and community mentions).

Other referenced providers / tech

Anthropic (Claude), Google Gemini, OpenAI ChatGPT, Antigravity, Banana.dev, Parallel AI, Aquavoice / Supervoice (STT), Obsidian (vault), VS Code plugins, Docker, Discourse, Synology (backups), Opus 4.6 (audio model).