every tool you need. none of the bloat.

built in rust. opens instantly, idles at <1% cpu, no electron tax.

p99 under 30ms, trained on your codebase — not the public internet.

predicts your next edit, not just your next token. accept with a single keystroke.

select text in the terminal, get an ai rewrite — without leaving the shell.

indexes a 10k-file monorepo in 2 seconds. incremental updates in milliseconds.

100% extension compatibility. bring your themes, snippets, and language servers.

split, tab, and task-integrated terminals — first class, not an afterthought.

set breakpoints, step through code, inspect state. full debug adapter protocol.

jupyter-style notebook surface for python, js, and more.

delegate multi-file tasks to an autonomous agent. it plans, edits, runs tests, opens a pr.

write, ask, plan, debug, secure, minimal, architect, refactor, performance — each with its own tools and model.

the agent proposes a plan, you approve before any edits land. trust, with a paper trail.

ask 'where do we handle stripe webhooks?' — get a cited answer in two seconds.

watch the agent reason in real time. thinking blocks, shimmer, live tool output.

every conversation saved, searchable, and shareable with your team.

favorite models, swap mid-thread, see cost per call. claude, gpt, gemini, local — your choice.

the right model for each message, automatically. cheap for trivia, smart for refactors.

the agent reads and writes vault docs and atlas data directly. no copy-paste prompting.

walks rust, typescript, python, go. finds crates, packages, entry points, and the edges between them.

read, edit, search, run, test, browse — every tool the agent needs, sandboxed and audited.

on by default. the agent can navigate, click, fill forms, and screenshot — to test what it ships.

the agent runs your test suite, reads failures, and iterates until green.

browse and install model context protocol servers in-app. no json wrangling.

install templated agents — qa, security, perf — from the registry.

plug in third-party agent servers. hellocode-agent, claude code, gemini cli, more.

lock down exactly which tools each profile can call. secure by default.

yaml-defined multi-step agent pipelines. 9 step kinds: prompt, tool, branch, loop, and more.

ships with feature-delivery.yaml and other starters. fork and adapt.

browse, run, and share team playbooks from the bottom dock.

/plan, /compact, /doctor, /init-rules, /help — fly through the editor without the mouse.

kick off a long-running workflow from any thread. monitor in the observe panel.

stage, commit, branch, push — native, no terminal needed.

run parallel agent sessions in isolated worktrees. ship three features at once.

share a live agent thread with a teammate. pair-program with ai in the loop.

see who's editing what, in real time.

lint, type, and build diagnostics piped straight to the agent so it can self-correct.

per-tool, per-thread, per-user token and dollar aggregation. no surprise bills.

every tool call written to a local audit.db. portable, queryable, yours.

live view of every tool call, every model request, every error. the glass cockpit.

saml sso, role-based access, scim provisioning. compliance teams approve in days.

run hellocode's hosted llm proxy, or point at your own vpc / on-device models.

limited-scope guest seats for contractors and reviewers.

encrypted local secrets store. api keys never leave your machine unless you say so.

roll out new agent capabilities to a subset of your team first.