Act
Inspect files, edit code, run commands, and work through explicit tool contracts.
Algo CLI combines direct tool use, durable context, routed agents, and explicit verification— without making a website or hosted account part of the runtime.
$ git clone https://github.com/Seabass-up/Algo-cli.git && cd Algo-cli
$ pipx install .
$ algo-cli --oneshot --json --approval-mode auto
“fix the failing parser test and verify the result”
◆ inspect src/parser.py · tests/test_parser.py
◆ edit src/parser.py +4 −2
✓ verify pytest 148 passed
✓ evidence tracked diff · file scope intact
done complete · verified delta
The product is organized around a simple operating loop: act deliberately, retain useful context, route only when complexity earns it, and verify before claiming completion.
Inspect files, edit code, run commands, and work through explicit tool contracts.
Carry durable context, lessons, and reusable skills across local sessions.
Keep ordinary work direct. Send larger tasks through bounded agent pipelines.
Ground completion in tests, file evidence, Git state, and fail-closed policies.
Algo CLI keeps simple tasks simple while exposing structured paths for larger, higher-risk work.
Read-only specialists gather evidence in parallel. One integration pipeline owns writes, review, and verification.
Memories, lessons, skills, and code retrieval have visible controls and privacy-safe defaults.
Use packaged public docs by default. Add local agent stores only when you explicitly opt in.
Algo CLI ranked first in our reported 27-run local comparison. That is a scoped result—not independently reproduced evidence or a universal superiority claim.
Inspect the methodology →