Tinyclaw
An open-source autonomous AI companion that learns your habits and drives your desktop, browser and APIs for you.
Tiny Claw is worth watching, not yet worth depending on. The vendor's own homepage says it is under heavy development and that running it now may spoil the intended experience, which tells you where the project stands. What is genuinely interesting is the engineering: an eight-dimension routing classifier to cut model costs, a three-layer memory with temporal decay, and a delegation system with self-improving role templates and a pub/sub bus — more than you get from a thin chat wrapper. Against that, there is no documentation or integration surface captured here, and the agent reads your screen, so reliability and privacy are real questions. If you build agents and enjoy rough edges, clone
Verified 1d ago · liveness 59/100 · cite: rightaichoice.com/tools/tinyclaw
- Developers comfortable reading source code to understand a tool
- Power users who want a self-hosted, extensible agent
- Agent-builders interested in memory, routing and delegation design
- Early adopters tracking autonomous-agent projects
- Teams that need a production-ready tool on a critical path
- Non-technical users who want plug-and-play setup
- Enterprises requiring compliance guarantees and data sovereignty
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Tiny Claw if you need an agent you can put in front of a team or a client this quarter, since the project itself says it is under heavy development and not yet ready.
Model routing cuts LLM spend on simple queries but complex tasks still route to stronger, more expensive models, so heavy autonomous runs can cost more than a chat assistant would.
Tiny Claw is a GPL-3.0 open-source project, so what you pay is model usage and your own setup time rather than a seat licence. Cost of ownership scales with how often you route complex work to stronger models and how many sub-agents a delegation spawns. That puts it well below subscription-priced hosted agents on licence cost, with the trade-off that you carry the engineering effort yourself.
In short
Tinyclaw — An open-source autonomous AI companion that learns your habits and drives your desktop, browser and APIs for you. Best for Developers comfortable reading source code to understand a tool, Power users who want a self-hosted, extensible agent, Agent-builders interested in memory, routing and delegation design. Free to start; paid plans from $20/mo.
What people actually say about Tinyclaw — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
3 mentions across 1 source (Hacker News) · researched Jul 3, 2026.
Average across the 1 source that answered — each source counts once, not each post.
- +Automates complex multi-step tasks across web, desktop, and APIs.
- +Computer Use paradigm allows direct GUI interaction like a human.
- +Includes human-in-the-loop checkpoints for critical steps.
- +Error recovery and self-correction aim to reduce failure rates.
- +Long-term memory across sessions for persistent context.
- −Too early-stage with minimal community reviews or validation.
- −Described as a wrapper around Claude Code, limiting uniqueness.
- −No integrations listed, requiring manual setup for most services.
- −Very limited public data: only one HN post directly about it.
- −Potential reliability issues due to incomplete error handling.
- • Claude Code API usage may incur additional charges
- • Premium features may appear later without notice
Viability Score
How well maintained and how widely used is Tinyclaw? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Autonomous GUI-level computer use (clicks, typing, screen reading)
- Dark-themed web chat UI with real-time SSE streaming
- Typing indicators and inline delegation event cards
- Active agents sidebar for watching sub-agents run
- Three-layer adaptive memory (episodic, semantic FTS5, temporal decay)
- Four-layer context compaction (pre-compression, dedup, LLM summarisation, tiered summaries)
- Eight-dimension query classifier routing cheap vs. strong models
- Self-improving behavioural pattern detection
- Delegation system with autonomous sub-agent orchestration
- Blackboard collaboration and adaptive timeouts for sub-agents
- Inter-agent pub/sub event bus with wildcard subscriptions
- Plugin architecture for channels, providers and tools
- Heartware personality engine (SOUL.md, IDENTITY.md)
- SHIELD.md runtime anti-malware enforcement engine
- Five-layer security (path sandbox, content validation, audit log, auto-backup, rate limiting)
About Tinyclaw
Tiny Claw is an open-source (GPL-3.0) autonomous AI companion built from scratch by Waren Gonzaga, currently in active development with no official release yet. It runs a dark-themed web chat UI with real-time SSE streaming, typing indicators and an active-agents sidebar, and it automates multi-step work by interacting with your machine at the GUI level as well as through files, terminal commands and APIs. Under the hood it keeps a three-layer memory (episodic, semantic with FTS5 search, and temporal decay) so it decides what to remember and what to let fade, plus a four-layer context compactor with rule-based pre-compression, deduplication, LLM summarisation and tiered summaries. An eight-dimension query classifier routes simple questions to cheaper models and complex ones to stronger models to keep LLM spend down, and a plugin architecture keeps the core tiny while channels, providers and tools are added separately. A personality engine (SOUL.md, IDENTITY.md) gives the agent a persistent character rather than a neutral assistant voice, and a delegation system lets it spin up sub-agents with blackboard collaboration over a pub/sub event bus. It is built on Bun for fast startup and a small footprint. Tiny Claw is aimed at developers, researchers and power users who want a persistent, self-hosted agent they can shape — not at teams that need a finished, production-ready product.
Behind the Verdict
The most useful thing about Tiny Claw is that it is not a prompt wrapper. The vendor describes a system built from scratch on Bun, with three layers of memory (episodic, semantic FTS5, temporal decay), a four-layer context compaction pipeline, an eight-dimension query classifier for cost-aware model routing, and a delegation system for autonomous sub-agents with self-improving role templates, blackboard collaboration and adaptive timeouts. Those are engineering decisions that shape day-to-day behaviour: memory decay means the agent is supposed to forget stale context rather than drown in it, and compaction is what lets long sessions stay coherent. The plugin architecture (channels, providers, tools all pluggable) and the pub/sub event bus are the right shapes if you intend to extend it yourself. The personality layer is the part buyers will either love or bounce off. SOUL.md and IDENTITY.md give the agent a persistent character, and the project is explicit that it is positioning the agent as a companion rather than a tool. That is a deliberate design bet, not a bug, but it means Tiny Claw is a poor fit if what you want is an impersonal automation runner. Security gets more attention than most early-stage agents: a five-layer model covering path sandboxing, content validation, audit logging, auto-backup and rate limiting, plus a runtime SHIELD.md enforcement engine described as anti-malware protection with threat parsing and pattern matching. Treat those as vendor claims you should verify yourself, because the scrape here is homepage copy only. The honest caveats are significant. The project states plainly that it is under heavy development and not yet ready, and the homepage asks you to follow for launch updates rather than install it. No documentation hub, changelog or integration catalogue was reachable in this pass, so anything you read about how it plugs into your stack is unverified. Beyond that, autonomy at the GUI level means the agent reads your screen, which is a categorically different privacy posture from an API-only assistant — and autonomous UI control is inherently brittle against apps that change layout. Where it fits: developers who want a self-hosted, hackable agent and are comfortable reading source to learn how it works. Where it does not: anyone who needs audit-ready behaviour, predictable uptime, or a tool a non-technical colleague can adopt next week. Watch the repo, but do not put it on a critical path.
Researching Tinyclaw? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Tinyclaw actually fits — and what changes day-one when you adopt it.
You ask Tiny Claw to clone a repo, install dependencies and run the test suite, watching progress in the web UI as it works through the terminal.
Outcome: The five-layer security model logs actions to an audit trail and backs up files, so a failed install step is recoverable rather than destructive.
You run a long research session that pulls from multiple web sources across hours, relying on the four-layer context compactor to keep the session coherent.
Outcome: Tiered summaries preserve earlier findings while temporal decay drops context that no longer matters, so the session does not collapse under its own history.
You add a channel and a tool as plugins on top of the minimal core, so the agent fits your own workflow instead of a vendor-defined one.
Outcome: Your additions sit alongside inter-agent pub/sub, letting custom tools communicate with delegated sub-agents.
Use Cases
- Drive a multi-step dev workflow across VS Code and the terminal without hand-holding each step.
- Run long research sessions and let memory decay drop stale context instead of re-reading everything.
- Orchestrate sub-agents on a shared blackboard for tasks that split into parallel workstreams.
- Build custom channels and tools as plugins on top of a deliberately small core.
- Give the agent a persistent persona for companion-style interactions rather than one-shot commands.
Models Under the Hood
as of 2026-09-09
Limitations
- The vendor states Tiny Claw is under heavy development and explicitly says running it now may spoil the experience they are building, so it is not positioned as ready for real work.
- No documentation hub, changelog or integration catalogue was reachable in this pass, so the surrounding ecosystem is unverified.
- Autonomy at the GUI level means the agent reads your screen and can be brittle against apps that change layout.
- A three-layer memory model with temporal decay and a four-layer compaction pipeline are ambitious designs whose real behaviour you should test yourself rather than take from homepage copy.
as of 2026-10-08
Verification history
We have re-verified Tinyclaw 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Tinyclaw's pricing actually pencils out — and where peers do it cheaper.
Tiny Claw is a GPL-3.0 open-source project, so what you pay is model usage and your own setup time rather than a seat licence. Cost of ownership scales with how often you route complex work to stronger models and how many sub-agents a delegation spawns. That puts it well below subscription-priced hosted agents on licence cost, with the trade-off that you carry the engineering effort yourself.
Setup time & first value
How long it actually takes to get something useful out of Tinyclaw — broken out by persona, not the marketing-page minute.
Developers: budget an afternoon for a first working run, since you are cloning a GPL-3.0 repo built on Bun and there is no reachable getting-started documentation in this pass. Power users: expect longer, because plugin wiring for channels, providers and tools is source-level work. Researchers: the fastest path to value is a single scripted task rather than a general-purpose setup.
Switching to or from Tinyclaw
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From AutoGPT: port your task definitions into Tiny Claw's delegation system, where sub-agents share a blackboard and use adaptive timeouts instead of fixed loops.
- →From a hosted chat assistant: move long-lived context into the three-layer memory so episodes and entities persist across sessions rather than living in one thread.
- ↗To AutoGPT: export your recurring task descriptions, since Tiny Claw's plugin and delegation concepts do not map one-to-one onto AutoGPT's agent format.
- ↗To a hosted agent platform: expect to rebuild screen-level automations as API calls, because GUI control does not transfer to platforms without computer-use.
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Tinyclaw”, and we withheld 6: 6 could not be judged, because “Tinyclaw” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Tinyclaw.
Official links
Tools that pair well with Tinyclaw
Common stack mates teams adopt alongside Tinyclaw, with the specific reason each pairing earns its keep.
Zhipu GLM
Zhipu GLM (Z.ai) ships the open-weights GLM-5.3 family, full-modality MaaS APIs, and autonomous agents like AutoGLM and GLM-PC.
Chrome DevTools MCP
Open-source MCP server that gives coding agents live Chrome DevTools access for debugging, automation, and performance traces.
Open Interpreter
Open Interpreter is an open-source terminal agent that turns plain-English requests into real file edits and shell commands on your machine.
Featured Head-to-Head Comparisons
Tinyclaw vs Presto Voice
Presto Voice and Tinyclaw are fundamentally different tools for entirely separate domains. Presto Voice is a specialized QSR drive-thru voice AI platform focused on revenue lift and operational efficiency, validated by chains like Dairy Queen. Tinyclaw is a general-purpose autonomous AI agent for developers and power users to automate complex workflows across web, desktop, and APIs. The choice is clear: if you run a QSR chain, go with Presto Voice; if you want a personal AI companion to automate tasks, pick Tinyclaw.
Tinyclaw vs Locus Robotics
Locus Robotics and Tinyclaw serve fundamentally different domains: physical warehouse automation vs. digital AI agent tasks. For logistics operations needing 2-3x productivity gains in picking/packing, Locus is the proven enterprise choice. Tinyclaw suits developers and power users who want an AI assistant to automate multi-step computer tasks across browsers, desktop apps, and APIs. Choose based on whether your problem is tangible goods movement or digital drudgery.
Tinyclaw vs Truleo
Choose Truleo if you're a law enforcement agency needing to unearth leads from siloed data and automate report writing. Choose Tinyclaw if you're a developer or power user wanting an autonomous agent to control desktop apps and browsers. They serve entirely different worlds.
Alternatives to Tinyclaw
View all
Zhipu GLM
Zhipu GLM (Z.ai) ships the open-weights GLM-5.3 family, full-modality MaaS APIs, and autonomous agents like AutoGLM and GLM-PC.
Chrome DevTools MCP
Open-source MCP server that gives coding agents live Chrome DevTools access for debugging, automation, and performance traces.
Open Interpreter
Open Interpreter is an open-source terminal agent that turns plain-English requests into real file edits and shell commands on your machine.
Frequently Asked Questions
Best-of guides
Used Tinyclaw? Help shape our editorial sentiment research.