local-first / open source


Chusquis.

HelloChusquis brings conversation, tools, memory, web access, code execution, and integrations into one developer-controlled workspace.

Explore the repository ↗Get started ↓
Built for people who want automation with visibility.
hellochusquis / local● ready
you › Inspect this repository and find the riskiest area.
agent › I’ll map the project first, then trace the highest-impact paths.

↳ shell find . -maxdepth 2 -type f
✓ 247 files found · 0.08s
↳ memory recalled 4 relevant decisions ↳ web checked current dependency advisories ✓ Risk map ready. 3 findings, 1 needs approval.
you ›
model: configured locallymemory: on  ·  session: isolated
Launching 31·01·2027 on Uneed — v5.4 is live, launch day is next.Notify me ↗Star on GitHub ↗
0+curated knowledge files
0+provider configurations
0service integrations
MITopen source license
the useful middle layer

Close to the work. Clear about the risk.

Most AI tools make you leave the place where the work happens. HelloChusquis keeps the agent close—and makes every important step visible.

01 / context

Remember what matters

Bounded history, searchable sessions, summaries, and local SQLite memory keep context useful without turning every prompt into a transcript dump.

01 / context

Memory, with boundaries.

Searchable sessionsRecall before each turnPin or forget factsLocal SQLite storage
Inspect the layer →
02 / reach

Bring your tools with you

Shell, code, files, browser, web search, and integrations live inside one conversation you can inspect as it unfolds.

02 / reach

One conversation. More reach.

Shell and bounded codeFiles and browser tasksCurrent web researchService integrations
Explore the tools →
03 / control

Automate with visibility

Approval gates, isolated HTTP sessions, protected credentials, and audit records keep high-impact operations accountable.

03 / control

Autonomy you can audit.

Approval gatesIsolated API sessionsProtected credentialsPersistent audit records
See the safeguards →
one surface, many ways in

Start where you are. Keep going.

Use the same agent from the terminal, authenticated web interface, REST API, or Telegram. Your workspace stays coherent as the surface changes.

01

Terminal

Interactive work, planning, tools, and fast feedback.

02

Web interface

Streaming chat, voice controls, export, themes, and approvals.

03

REST API

Applications, automation, isolated sessions, and service integrations.

from zero to useful

A calm first command.

Install it, configure a provider, and start asking for concrete work. No new workspace to learn.

quickstart.sh
# install
python -m pip install hellochusquis

# configure a provider
hellochusquis config

# start chatting
hellochusquis
recent project signals

The latest updates, in the numbers.

Two snapshots already shipped with the repository: knowledge retrieval lifts factual accuracy, while v5.2 cuts the slow paths down.

Factual accuracy improves with MDknowledge across model sizes
MDknowledge improves recall35-question factual probe
Latency before and after v5.2 on key operations
v5.2 removes waitingseconds, log scale
open source / developer controlled

Make the agent part of the place you already work.

Read the code, run it locally, and shape the tools around your own workflows.

Go to GitHub ↗