Keep the work. Focus the context.
An MIT-licensed coding and research agent with a persistent Python workspace, parallel workers, and source-backed context. Developed by Synerise, built on Prime Agent.
30 Python tasks × two profiles: GPT-6.1 Sol high and GPT-6 Astra medium, with best-of-two selection for both tools. Time is mean task-run duration; costs estimate the captured API usage of selected runs. Full results and method.
Install · Use it · How it works · Documentation
On macOS or Linux, the recommended installer prepares missing prerequisites and managed Python before activating the CLI:
curl -fsSL https://github.com/BaseModelAI/base-context/releases/latest/download/install.sh | bashRun the final PATH command it prints, then:
cd /path/to/project
base-contextUse /login for your provider, /model for your model, and /effort for its reasoning level. Selection and authentication are explicit; Base Context uses its own settings and credentials.
npm alternative: if you already manage Node.js/npm, use Node.js 22.12+ on 22.x, or 23.3+. In Bash/Zsh:
npm install -g @ponythewhite/base-context
BASE_CONTEXT_INSTALL_UV=1 base-contextThe first normal launch prepares managed Python. The flag permits installing missing uv; you do not need a manual Python installation. Use npm instead of, not in addition to, the installer above.
Windows, source builds, updates, rollback, and uninstall.
Start in your repository and ask for an outcome. Put project rules in AGENTS.md. You do not need to write Python yourself.
| Command | Purpose |
|---|---|
/goal <objective> |
Explicitly start a persistent goal; manage it with status, pause, resume, or clear |
/agents [N] |
Show or save the live-subagent cap, default four; lowering it does not stop workers |
/btw <question> |
Ask a tool-free side question without adding the conversation to the main session; Esc returns |
/context, /usage |
Inspect context, captured usage, cost estimates, and separate goal-budget scope |
/compact, /refine |
Summarize context or refine saved harness advice |
base-context agents |
Open the agents view |
base-context attach <agent> |
Reattach to a resident agent |
base-context shutdown |
Stop all agents and background services |
For independent work, ask:
Delegate the API review and documentation update to separate workers.
Keep working on the parser fix, then integrate their replies.
Normal interactive workers can continue after the terminal detaches. Quickstart · Full usage.
- Persistent Python: keep variables, parsed data, commands, and skills available across tool calls.
- Explicit working context: TaskFrames carry selected task state; ViewUnits keep required messages together; indexed recovery retrieves retained public evidence.
- Parallel children:
await rlm(...)returns an admission handle. Results arrive through messages or files, while independent work can continue.
Base Context inherits the Python-first programming model from Prime Agent. Its focus is the source-backed working-set architecture around that model. Bring that workflow to your next coding or research task.
Summaries are not lossless, retrieval is bounded, and Python restoration is best-effort. Generated code runs with your user permissions; workers are not a security sandbox. Use an external sandbox for untrusted work. Retained sessions can contain sensitive data; compaction does not delete it.
Architecture, comparison charts, and design details · Context management · SDK · All docs.
MIT. Upstream notices are preserved in NOTICE.
Thank you to Prime Intellect for the excellent Python-first agent foundation, and to Mario Zechner's Pi, on which Prime Agent builds. Base Context is an independent fork, not an official Prime Intellect release.