clawops
Get Started
Fine-Tuned

Engineered to work well out of the box.

Every ClawOps bot ships with a memory system, a boot sequence, a document search engine, and rate limit protection. Not configuration options — defaults that just work.

session boot
# ClawOps agent initialising...
→ loading your profile
→ reading recent context (today + yesterday)
→ loading long-term memory — 14 facts
→ applying 6 learned rules
→ qmd search: "api spec" → 3 chunks retrieved
→ context ready. token budget: 7,400 remaining.
Identity
Memory
QMD
Ready

Your workspace. Pre-configured.

Five markdown files, written at provision time. The bot reads them at the start of every session. Edit any of them directly or ask your bot to update them.

SOUL.md

Persona and platform awareness

Tells the bot who it is and how to behave. Pre-seeded with VPS awareness, the files convention, and platform knowledge. Edit anytime to change tone, style, or focus.

> You live on a dedicated Linux VPS, always on > Save all output to ~/files/ — user sees it in Dashboard > Be concise. Match user energy. Skip filler.
AGENTS.md

Boot sequence and write discipline

Loaded every session. Tells the bot what to read at startup, when to write context back, and how to handle handovers before model switches.

# Boot: load profile, recent context, long-term memory # Write: log after tasks, handover before switches # Never lose track mid-session
USER.md

Your profile

Seeded at provision time with your name and Telegram handle. Add your timezone, job, preferred tools, and working style. The bot reads this first, every session.

> Name: Alex > Telegram: @alexdev > Stack: TypeScript, Convex, React > Timezone: CET (Berlin)
MEMORY.md

Long-term memory

Curated facts that carry across all conversations. The bot updates it during quiet moments, never mid-task. You can edit, expand, or prune it directly.

- Prefers short answers unless asked to elaborate - Working on ClawOps SaaS (April 2026) - Stack: TypeScript, Convex, React
LEARNINGS.md

Rules from experience

Pre-seeded with known rules so your bot doesn't have to learn things the hard way. Grows over time — when the bot corrects a mistake, the rule gets written here.

- Always save output to ~/files/, not workspace - Check memory before answering questions about the user

Memory that actually persists.

Two tiers of memory working together, plus automatic compaction so nothing important is lost when the context window fills up.

Short-term: daily logs

After every significant task, the bot appends a note to today's log. Yesterday's and today's logs are auto-loaded at session start.

memory/YYYY-MM-DD.md

Long-term: curated memory

MEMORY.md holds the facts that matter across months. Updated during quiet review moments, never mid-task. Mistakes go to LEARNINGS.md.

MEMORY.md + LEARNINGS.md

Compaction flush

When the context window starts to fill up, important context is automatically written to the daily log before the reset.

auto-flush before context reset
Built-in skill

QMD saves 80-90% of document tokens.

Every bot ships with QMD, an open-source local search engine. Instead of loading entire files into context (expensive), your bot searches locally first and retrieves only the relevant chunks.

Keyword search BM25 ranking, same algorithm as search engines, instant
Vector search Semantic similarity using local embeddings, no API calls
Indexes your files Documents in ~/files/ are searchable by default
100-page document query
Without QMD ~40,000 tokens · $0.12
With QMD ~3,000 tokens · $0.009
Monthly cost, 100 queries
$12
Without
$0.90
With QMD

Not your average chatbot.

Generic AI tools reset every session. ClawOps ships with memory, recall, and reliability built in.

Feature ClawOps

Persistent memory across sessions

MEMORY.md + daily logs loaded at every session start

Yes

Context survives window resets

Compaction flush at 4,000 tokens writes to disk before reset

Yes

Document recall at 80-90% fewer tokens

QMD retrieves relevant chunks instead of loading full files

Yes

Rate limit failover, silently

Primary + 2 backup models per stack, you never notice

Yes

Context wiped on each new chat

Unlike cloud AI apps, your workspace carries context forward

No

Settings reset when platform updates

Your VPS workspace is yours, platform updates never touch it

No

Self-healing

Roadmap

Your bot will run nightly health checks automatically — compacting old logs into long-term memory, re-indexing new files into QMD, verifying its own config, and alerting you to anything that needs attention.

Ready to go on day one.

Memory, recall, fallbacks, and a boot sequence that loads your context automatically. All included, nothing to configure.

Start now, $10/month