What it does
5 features
PicoClaw · Self-hosted personal agent
Product and access
PicoClaw is listed as a separate product profile.
Core features
5 features from provider sources · checked on 16 October 2026. Availability and limits are listed below.
Usable AI models
Open choice
You add providers with your own key to the model list, plus OpenRouter, LiteLLM or an OpenAI-compatible endpoint via api_base. Models run locally through Ollama, LM Studio or vLLM.
Model families named
- OpenAI (GPT)
- Anthropic (Claude)
- Google (Gemini)
- DeepSeek
- Alibaba (Qwen)
- Moonshot (Kimi)
- MiniMax
- Z.ai (GLM)
- Mistral
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Yes
The rating covers the generally available main access. Documented · Sources accessed on
Does it fit my task?
2 practical tasks · 3 requirements and limits
A good fit if …
- you want to run an agent as a single Go binary on small ARM, RISC-V or MIPS machines
- you want to use your own model keys or local models through Ollama or vLLM
- you want to reach it through Telegram, Discord, Slack or another of the 19-plus chat channels
Less suitable if …
- you need a production-ready setup; the README warns against use before version 1.0
- you need organisation-level administrator approval; the reviewed pages describe none
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Set up a small computer
- Your input
- A supported computer and model access.
- Possible result
- An initial chat through the configured gateway.
- What to check
- Check content, permissions and actions taken.
Analyze an image
- Your input
- A non-sensitive image and a compatible model.
- Possible result
- An image description for your review.
- What to check
- Record time spent, corrections and costs.
Set up a small computer.
Your possible prompt
Organize these shared documents by topic. Create an overview with deadlines and next steps. Prepare draft replies for my approval.
Adjust the task, data and permissions.
Requirements and limits
Can I use it?
With requirements
Europe & regional availability
Installation instructions and model prerequisites reviewed. The assessment covers the local operating path described; complete country and feature availability of all connected services remains open.
Reviewed scope: Local software and documented self-hosting; external services assessed separately
Reviewed on:
With requirements
Self-hosting with own model access
PicoClaw offers local binaries and Docker operation. The documented startup configures an LLM API key and optionally a chat channel; check their country availability separately.
Region, reason and source
Region covered: Europe: own computer or server; model services are country-specific
Documented reason: Technical assessment based on installation and provider selection. These sources do not establish a legal cause for regional restrictions.
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Self-run PicoClaw instance (Go binary) with your own model keys and chat channels
Sources reviewed:
Show all 3 points
Documented
Account used
You enter model access as API keys, which are loaded from a .security.yml file kept apart from the main configuration. The Telegram channel uses a bot token, and some channels such as WhatsApp and Weixin can be paired by QR scan. A few model providers use OAuth instead.
Evidence · EvidenceOpen in reviewed sources
Admin approval
The reviewed pages mention no organisation or administrator approval. The configuration page shows a list of allowed user IDs (allow_from) for Telegram but describes no access rules for other channels.
EvidenceDocumented
Data access & team sharing
Commands and file access of the exec tool are limited to the workspace by default (restrict_to_workspace is on by default). Conversations are stored under sessions/ and long-term memory under memory/. session.dimensions controls how much memory chats, users and threads share.
Evidence
Memory and conversation continuity
3 of 6 criteria documented
A long-term file and daily notes provide file-based context.
Workspace memory in current source code
Sources accessed:
Documentation by criterion
StorageDocumented
Storage and scope
The workspace contains memory/MEMORY.md and daily notes under memory/YYYYMM/YYYYMMDD.md; writes are atomic.
SelectionPartly documented
Selection and timing
The code provides writing and appending; specific automatic fact selection and its triggers are not defined there.
Edit & deleteDocumented
Inspect, edit and delete
Local memory files can be read, edited and deleted; workspace-file changes are considered on subsequent requests.
Access boundariesPartly documented
Agent and channel boundaries
MemoryStore binds to the configured workspace. Fine-grained user and channel separation of memory is not described in the module.
RetrievalDocumented
Retrieval
GetMemoryContext combines the long-term file with daily notes from the last three days; it contains no vector retrieval.
New conversationsPartly documented
Session reset and compaction
Memory files are separate from saved sessions. Reset and compaction consequences for memory are unspecified in the inspected sources.
Cost and effort
Prices as of 16 October 2026
Prices and included usage
Pricing checked
Open source; model and operating costs are separate.
- Total cost depends on access, usage and configuration.
Setup
Download a matching release; configure a model provider and chat channel.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Self-run gateway on your own machine or server
Sources reviewed:
Editorial assessment
You run and maintain PicoClaw yourself; state lives in files, while failure handling and recovery are barely documented. The README advises against production use before version 1.0.
Show all 6 points
Documented
Files and working state
The workspace (default ~/.picoclaw/workspace) holds sessions, memory, state, cron and skills. Scheduled jobs are stored in <workspace>/cron/jobs.json. Changes to the files AGENT.md, SOUL.md, USER.md and MEMORY.md are detected at runtime, so no restart is needed.
Evidence · EvidenceOpen in reviewed sources
Sign-ins
The reviewed pages do not say how long bot tokens or model keys stay valid or what happens when they expire. That depends on the respective provider.
EvidenceDocumented
Starting further runs
Further runs start from a chat message, a cron job or the heartbeat. The cron tool supports one-off, interval and cron-expression schedules; the command line supports recurring jobs only. By default the heartbeat reads HEARTBEAT.md every 30 minutes (configurable) and can be switched off.
Evidence · EvidenceDocumented with conditions
Human intervention
Cron jobs with shell commands are gated: remote channels need an allowlist (empty by default), and with allow_command set to false a command job runs only with command_confirm. The reviewed pages describe no general approval of every action; the README mentions approval hooks in the hook system.
Evidence · Evidence · EvidenceOpen in reviewed sources
Detecting errors
You set verbosity through gateway.log_level (default warn). When command execution is disabled, existing cron commands report an error when they fire. The pages give no overall error overview; the troubleshooting page covers only two model errors.
Evidence · Evidence · EvidenceOpen in reviewed sources
Resuming after failures
Cron jobs and sessions are files in the workspace; one-off jobs are deleted after running, recurring ones remain. The reviewed pages do not describe retries, catching up on missed runs or resuming after a restart. Command jobs have a timeout (default five minutes).
Evidence · Evidence
Evidence and open questions
1 source · 2 open points
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Requirements and limits: not yet confirmed
Sources and evidence
Provider sources reviewed; examples are editorial. No hands-on tests.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
Publishes PicoClaw and the linked product information.
- Operator
- Sipeed
- Organization
- Open-source project
- Website
- github.com
Changes to this profile 1
Added PicoClaw with capabilities, sources, pricing and open European availability questions.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.