Use Colossus¶
Start a task in the terminal or an editor, return to it later, and bring in the information it needs. These guides show the user-facing workflows; each keeps the same Colossus model routing, tools, approvals, and session history.
Connect a model¶
-
Choose a provider
Connect a subscription, API key, local model, or compatible endpoint.
-
Local models
Work with a model server you run yourself.
Choose where to work¶
-
Terminal UI
Talk through a task, watch tools run, and answer approvals in the terminal.
-
Agent runs
Run one prompt from the shell or capture JSON for a script.
-
ACP in editors
Use Colossus as a local agent in an ACP-compatible editor.
Keep work moving¶
-
Sessions
Find a conversation and continue it from the CLI or terminal UI.
-
Context
Read the model input budget and manage long-session snapshots.
-
Planning
Review a durable plan, then run it once or through a bounded goal loop.
-
Goals and subagents
Iterate on a longer objective or delegate bounded work.
Find and keep information¶
-
Search
Run a web search, understand its results, and see how agent and research routes use it.
-
Deep research
Produce a durable cited report from repository, web, or MCP evidence.
-
Memories
Save reusable context for later work.
-
Decisions
Record a binding workspace choice that future work should respect.
-
Tools
See the catalog of actions the agent can take in this workspace.
For skills, plugins, MCP servers, and workflows, see Extend Colossus.