The mental model shift
Most AI agent tools optimize for autonomy — give the agent access to everything and let it act on your behalf. OpenClaw connects to 50+ integrations, runs shell commands, browses the web, and ships with permissive defaults. The pitch is “your AI can do anything.” ollim-bot optimizes for context quality. The belief is that an assistant is only as useful as how well it understands you right now. Autonomy, features, integrations — they’re all secondary to whether the agent actually knows what’s going on in your life. This isn’t a feature gap — it’s a design choice. ollim-bot does less on purpose, because doing fewer things with deep personal context beats doing everything with shallow context.Read Design philosophy for the full reasoning behind
these tradeoffs.
Feature comparison
An honest mapping of what exists in typical AI agent tools vs ollim-bot. Blanks are intentional — not everything needs an equivalent.| Capability | OpenClaw / others | ollim-bot |
|---|---|---|
| Messaging platforms | WhatsApp, Telegram, Slack, Discord, 50+ | Discord DMs only |
| Persistent memory | Markdown files + database indexes, session-independent | Persistent sessions with compaction |
| Scheduling | Cron jobs + periodic heartbeats | Routines (contextual check-ins) + reminders with follow-ups |
| Background work | Autonomous execution | Background forks with ping budget |
| Proactive outreach | Varies by tool | Core feature — the bot reaches out to you on schedule |
| Multi-provider | Gmail integration | |
| Calendar | Multi-provider | Google Calendar integration |
| Tasks | Various providers | Google Tasks integration |
| Shell access | Direct host access by default (sandboxing optional) | Available, permission-gated per session |
| Web browsing | Built-in | Available, permission-gated per session |
| Plugin marketplace | Large community ecosystems | No marketplace — subagents, skills, hooks, and webhooks you configure yourself |
| Multi-user | Yes | Single-user only — by design |
| Data storage | Varies — Markdown + SQLite (OpenClaw), proprietary (commercial tools) | Markdown + JSONL files — no database dependency, git-trackable |
| Extensibility | Plugin SDKs, community skills | Subagents, skills, hooks, Discord tools, system prompt |
What you gain
Context that persists and compounds. ollim-bot maintains a single persistent session that picks up where you left off across restarts. As context grows, compaction summarizes older history to free space — so the bot retains the important parts without hitting context limits. Background tasks fork from the main context rather than starting fresh, so they inherit what’s already been discussed. Proactive check-ins, not just reactive responses. Most AI assistants wait for you to start a conversation. ollim-bot messages you on schedule — morning briefings, end-of-day reviews, task follow-ups. For ADHD, this is the difference between a tool you forget to open and an assistant that catches things before they slip. Notification management that respects your attention. The ping budget limits how often background tasks can notify you, so scheduled work doesn’t bury you in messages. The bot decides what’s worth pinging you about and what can wait. Transparent, auditable data. Routines and reminders are markdown files with YAML frontmatter — open one in any editor and see exactly what it does. Session logs are append-only JSONL. All data lives in~/.ollim-bot/, and most files are auto-committed to git for version history. No database indexes to rebuild, no migrations to run.
Smaller attack surface. Tools with large plugin marketplaces and permissive defaults
have a larger attack surface. ollim-bot’s tools are permission-gated, actions are visible
in your Discord DMs, and there’s no plugin marketplace where unreviewed code can run.
What you give up
What ollim-bot doesn’t do:- Platform choice. Discord only. No WhatsApp, Telegram, Slack, or SMS. If Discord isn’t where you spend your day, this isn’t the right tool.
- Broad automation. Shell, web search, and file tools exist but are permission-gated and denied by default. No browser automation, no “do anything” defaults. ollim-bot is a personal assistant, not an autonomous agent platform.
- Plugin ecosystems. No marketplace, no community repository. You extend it with subagents, skills, hooks, and webhooks — more control but more setup.
- Multi-provider integrations. Email, calendar, and tasks are Google-only. If you’re on Outlook or another provider, those integrations don’t exist yet.
- Multi-user support. One bot, one person. If you need a team assistant or shared workspace, look elsewhere.
Migration tips
There’s no automated migration path — memory formats, session stores, and configuration systems are too different across tools. But you can get up to speed faster than starting from scratch. You can import context from your previous assistant, recreate your automations, or set up new integrations.Import your context
Import your context
If your previous assistant stored memory as Markdown files (OpenClaw’s
MEMORY.md,
daily logs, etc.), drop them anywhere in ~/.ollim-bot/ (the bot’s
data directory) and ask the bot to read through them:“I put some files from my previous assistant in the data directory. Read through them, then check your docs to see what you can actually do with what you find — routines, reminders, integrations, whatever fits.”The bot reads your files, cross-references them against its own documentation, and suggests what it can set up for you. This isn’t a memory import — it’s a conversation where the bot gets to know you faster by reading what your last assistant learned and connecting it to what it can do.
Recreate your automations
Recreate your automations
ollim-bot can read its own documentation, so everything in these docs
directly helps the bot set you up. You don’t need to know the exact feature names —
describe what you had and let the bot figure out how it maps:
“I used to have a morning check-in that reviewed my calendar and tasks, plus reminders that followed up until I actually did the thing. Check your docs and set up whatever fits.”The bot reads the same routine, reminder, and scheduling docs you do. It knows the file formats, frontmatter options, and scheduling patterns — describe what you want and it creates the files. See real-world examples for common patterns.
Add new integrations
Add new integrations
ollim-bot already has shell access, file tools, web search, and code execution —
but the permission system denies unapproved tools by
default. New integrations (Google services, webhook
endpoints, structured tool APIs) follow established patterns in the codebase.
“I want to connect a new service. Check the extending docs and walk me through what’s possible — integrations, subagents, webhooks.”See adding integrations for the full procedure, and subagents for delegating specialized tasks.
ollim-bot is open source and contributions are welcome — especially new integrations and
MCP tool configurations. If you build something that follows the
design philosophy, open a PR.
Next steps
Quickstart
Get the bot running in 7 steps.
Design philosophy
The full reasoning behind these tradeoffs.
Routines
Set up proactive check-ins — the feature that makes the difference.
How it works
The agent loop, sessions, and fork model under the hood.