Request
Describe the outcome in your own words and choose the workspace context.
Review the retry logic, consult the relevant docs, and run the tests.
A self-hosted AI workspace
Go deeper.
Bring the work together.
Your models, tools and knowledge. Connected in a workspace you host.
CodeInspect a changeRun the relevant tests
SourcesRead project knowledgeKeep references attached
Plan✓ Review the scope○ Check the result
Illustrative workflowFollow the work,
not just the answer.
See the request, the tools used, and the result produced. Permissions and execution records help you decide what runs and what needs review.
Work
The root agent can use tools directly or delegate bounded tasks to specialists.
✓ Searchproject docs
✓ Readsource files
○ Runfocused checks
Review
Inspect artifacts, sources, decisions, and checks before accepting consequential work.
ChangesSources
ChecksExecution log
Different work, shared context
Move from idea
to useful output.
Octipus combines configured models with connected tools. What it can do depends on the models, permissions, and integrations you choose.
Code
Investigate a codebase, prepare a change, run focused checks, and leave the evidence ready for review.
Research
Gather from the web and your knowledge base, retain source references, and turn findings into a report or follow-up task.
Automation
Build scheduled or event-driven workflows with explicit stages, permissions, retries, and human input where it matters.
Models and control
Your models.
Your rules.
Run Octipus on your infrastructure and connect local, hosted, compatible, or CLI-backed models. External providers receive the data your configured work sends to them.
- Bring your modelsUse Ollama, major hosted providers, compatible endpoints, LiteLLM, or supported CLI agents.
- Set permissionsALLOW, ASK, and DENY rules govern built-in tool execution, with stricter behavior for unattended work.
- Work where you preferConnect through the web app, terminal, desktop client, supported messaging channels, voice, or MCP.
v0.3 alpha · building in public
Your work has depth.
Give it a workspace that does too.
Start with the guided setup, then connect the models and capabilities your work needs.