For the last decade, every team has stitched together the same four-piece kit: a chat assistant to think with, a coding agent to write with, a no-code builder to ship interfaces with, and an automation tool to wire it all together. Four bills. Four seat counts. Four sets of context that never quite talk to each other.
Shogo collapses all four into one open, self-evolving platform. Describe what you want — a dashboard, a workflow, an internal tool, a new feature in your repo — and a Shogo agent builds it, runs it, and keeps improving it. You own the code. The agents keep evolving it as your business changes.
Today we’re shipping Hoshi 1.0, the new intelligence layer underneath. It’s what finally makes a single agent good enough to do all four of those jobs at once.
The four-in-one platform, in plain terms
Every Shogo agent is the same agent. Same memory. Same tools. Same access to your data. What changes is which pillar of work it’s doing at any given moment:
- 01 · Your AI assistant. Ask it anything, like you would Claude or ChatGPT — except it can actually see your stack, your data, and your team. The conversation is the front door to every other pillar.
- 02 · Your coding agent. Ship features, not edits. Shogo writes, reviews, and lands code in your real repo — across languages, across files — without you babysitting the diff.
- 03 · Your app builder. Describe a tool and get a working one back: live screens, a real database, auth, and the routes wired up. No “export to code” wall later — the code is yours from day one.
- 04 · Your automation stack. 1000+ optional bridges into Slack, Notion, GitHub, Linear, Stripe, Salesforce, your warehouse, your inbox. Workflows that read, decide, and act — not just trigger and forward.
The interesting part isn’t any one of the four. It’s what happens when the boundary between them disappears. Asking a question (pillar 1) can turn into editing your repo (pillar 2). Editing your repo can spin up a new internal tool (pillar 3). The tool can quietly run on a schedule, pulling from Stripe and posting to Slack (pillar 4) — without you switching apps, re-explaining context, or paying four vendors.
Self-evolving, by design
Most AI tools today are frozen. You prompt them, they answer, and the work evaporates. Shogo agents are different: every workflow you build is a living artifact they can come back and improve. New integration available? They rewire. New edge case in your data? They handle it. New process on your team? They update the tool — and tell you what changed.
That only works if the model underneath is cheap enough to keep running. Frozen tools are frozen because $5-per-task economics force them to be. Hoshi 1.0 lands agent-shaped work at roughly $0.005 per task — two orders of magnitude lower — which is why the “self-evolving” claim stops being a slide and starts being how the product actually feels.
Open, model-agnostic, yours
Three things Shogo has always promised, and Hoshi 1.0 doesn’t change any of them:
- Open platform. The agents, the runtime, and the templates are open source. You can self-host the whole thing. There is no proprietary lock-in waiting at the end of your trial.
- Model-agnostic routing. Hoshi 1.0 joins a routing table that already includes Anthropic, OpenAI, Google, and local models. Pick the best model per task by score, cost, or privacy — your call, per workflow.
- You own the code. Whatever a Shogo agent builds for you — apps, workflows, schemas — lives in your repo, under your license. If you walked away from Shogo tomorrow, you’d walk away with working software, not a screenshot.
Proof it actually works
A four-in-one platform only earns the name if the model underneath is reliable enough to drive all four pillars. We ran Hoshi 1.0 against the two closest alternatives — OpenClaw + Claude Haiku 4.5 and OpenClaw + gpt-5.4-nano — on three independent, harness-neutral evals: one each for agentic tool use, structured output, and multi-language coding. Hoshi 1.0 wins all three, at a fraction of the cost.
| What it tests | Maps to pillar | Hoshi 1.0 | Best alternative |
|---|---|---|---|
| Agentic tool use (ClawBench) | Automation | 0.844 | 0.797 |
| Structured output (ClawEval)* | App builder | 88.4% | 84.7% |
| Multi-language coding (Aider) | Coding agent | 83 / 83 | 81 / 83 |
| Cost per task (Aider set) | All four | $0.005 | ~$0.06 |
That’s the floor under the platform. Full benchmark methodology, per-language splits, and Terminal-Bench numbers are in the Hoshi 1.0 technical report.
From the team
“We didn’t set out to build another model. We set out to collapse four categories of software into one platform that anyone — not just engineers — can use to ship real tools. Hoshi 1.0 is the piece that makes that single platform good enough to replace all four. Open, model-agnostic, and yours to own.”
— Guru Angisetty, CEO and Co-Founder of Shogo
“Self-evolving software has been the dream for a decade. The reason it never landed is that nobody built one agent that could think, code, build, and automate from the same memory. Shogo is that agent. Hoshi 1.0 is what lets it keep working — quietly, continuously — while you focus on what to build next.”
— Russell LaCour, CTO and Co-Founder of Shogo
About Shogo
Shogo replaces four categories of tool with one open-source platform: your AI assistant, your coding agent, your app builder, and your automation stack. Describe what you want, and a Shogo agent — powered by Hoshi 1.0 — builds it, runs it across 1000+ optional bridges, and keeps improving it as your business changes. You own the code. You pick the model. You walk away with working software.
Deploy your first agent in minutes on the free tier, or browse the templates to see what teams are shipping today.
* ClawEval reflects Shogo’s full agent on the ClawEval v2 leaderboard (Phase H, 1,220 checkpoints), scored with the lenient (uniform) scorer: 1,078 / 1,220 = 88.4%, #1 on the leaderboard. Figures shown for other models are their own published results.