The Bodega One Code Roadmap
This is what we shipped, what we are building right now, and what comes next. No vague promises. No “exciting things coming soon.” Just the actual list.
Beta is live now. Full launch coming later this year. See the full changelog for version-by-version history.
Shipped at Beta Launch
Everything below was in the product on day one of the open beta. It is all in the free download today, and every one of these capabilities stays in the free Personal tier at full launch. See the changelog for the full build log by version.
Editor
- Monaco editor
- Multi-file tabs with drag-to-reorder
- Inline streaming diff - character-level highlighting as agent writes
- Hunk-by-hunk diff review: accept or reject per code hunk
- FIM code completion as you type (like Copilot, fully local)
- LSP client: IntelliSense, diagnostics, go-to-definition
- Git integration with AI-generated commit messages and PR descriptions
Terminal
- Multiple terminal tabs with xterm.js full emulation
- AI terminal assist: command suggestions and error explanations
- Shell security: risk classification before execution
Chat Mode
- 26 built-in tools: file ops, shell, web search, memory, multi-agent
- Token-by-token streaming with 66x render reduction
- Research mode: web synthesis before responding
- Extended thinking support for compatible models
- Tool call display with expandable cards
- Tool approval cards: approve or reject before execution
- Ghost text prediction in the chat input (Tab to accept)
- Speech-to-text via Ollama Whisper (fully local, air-gap safe)
Autonomous Agent
- Autonomous coding agent with up to 16-iteration agentic loop
- Permission modes: Ask, Plan, and Act
- Multi-agent swarm: spawn parallel worker agents mid-task
- Repo map with PageRank file ranking
- KV cache optimization: 40-70% reuse per session
- Checkpoint system: per-tool file snapshots for rollback
Quality Enforcement Layer (QEL)
- Syntax and semantic checks after every file write
- Proof gates: compile-based verification for Go, Rust, Java, C#
- Structural verifier: catches stubs, incomplete code, hallucinated imports
- Learning service: tracks failures per tool-model to prevent repeat errors
Privacy and Security
- Air-gap mode with 9 enforcement layers - zero network egress
- Observation masker: strips credentials and PII before LLM calls
- SSRF protection: private IP blocking, non-HTTP protocol blocking
- Shell credential scan: catches leaked keys in terminal output
LLM and Providers
- 10+ provider presets: Ollama, LM Studio, vLLM, llama.cpp, OpenAI, Groq, and more
- BYOLLM: no bundled model, no markup, no usage tracking
- Hardware-aware model recommendations based on detected RAM and VRAM
- Provider auto-detection: scans for running LLM servers on startup
- Fallback routing: automatic failover on provider errors
- OpenAI-compatible API server on port 1337 (Bodega One Code as a provider)
Memory and Knowledge
- 4-layer memory: session context, project context, user preferences, long-term facts
- Knowledge base: add URLs or text as persistent context sources
- Skills system: YAML-defined custom skills with hot-reload (no restart needed)
- MCP server support: connect any MCP server with tool namespacing
- Per-session and per-project memory scoping
Onboarding and Settings
- First-run wizard: provider auto-detection, model download flow
- Guided tour: highlights key UI elements in both Chat and Code modes
- 60+ configurable settings keys
- Settings import/export as JSON
Shipped in the Open Beta
The beta opened in May 2026 and has shipped a tagged release most weeks since. These are the headline additions, each tagged with the version it landed in. The changelog has every release in full.
Agent Orchestration
- Fleet: background agent sessions in isolated git worktreesbeta.20
- Fleet Parallel: one task fanned out to multiple models, QEL scores the attemptsbeta.25
- Bodega Loops: scheduled agent tasks. Every run lands as a reviewable diff with its QEL scorebeta.27
- Run Tasks and the Runs Inbox: one-click dev servers, one surface for everything waiting on youbeta.28
- Custom agents: profiles whose tool allowlists can only narrow, never widenbeta.26
- Custom agent starter templates (Code Reviewer, Test Runner, Doc Checker) and per-agent run history with QEL score and tracebeta.29
- Routing rules: an ordered, first-match-wins table that routes each task, step, or file to the right model, or keeps it localbeta.29
- Verified Private Automation: point it at a task or a GitHub issue and it does the whole job on your machine, working in an isolated worktree, running full QEL, and opening a finished PRbeta.29
- Agent memory: it learns what worked on past tasks and from the tools you reject, on-device and scoped to you. In-context memory, not fine-tuningbeta.30
- Self-improving Loops: a scheduled Loop reads its own run history, carries forward what failed, and auto-pauses if its quality dropsbeta.30
- Lifecycle hooks: run your own shell command to lint, format, or block a tool before or after it runsbeta.31
- Named permission profiles: reusable rule sets that can only narrow what the agent may do, by tool or path globbeta.31
- Watch-mode comment triggers: act on inline // bodega: comments on save, through the same verified apply gate as every Loopbeta.31
- Smart approval: in Ask mode, low-risk read-only tool calls that match your request can auto-approve while write tools stay gated. The classifier runs last, after every security gate, so it only ever narrows what you are asked to approvebeta.31.3
- Author skills with /learn: point it at a workspace folder or a URL and it drafts a skill, shows a preview, and writes it only on approvalbeta.31.3
- Auto-skill capture: a creation task that passes QEL through a real multi-step workflow is saved as a local, per-user knowledge card and recalled on similar requestsbeta.31.3
- Bodega Mixture (Mixture-of-Agents): run several reference models in parallel on one turn and have an aggregator model synthesize the single answer, with an optional QEL quality gate. Off by default, air-gap awarebeta.31.6
Interoperability
- ACP host: Cursor, Claude Code, Gemini CLI, and Codex run inside the fleet, sandboxedbeta.25
- ACP server: bodega --acp lets Zed drive the Bodega agentbeta.26
- One-click rules import from Cursor, Copilot, Cline, Windsurf, and Continuebeta.28
- MCP per-server tool filtering, plus required servers that fail fast instead of silently degradingbeta.31
Verification (QEL)
- Execution proofs: QEL boots the generated server and probes the route the task asked forbeta.27
- Wider language coverage: vitest gates for JS/TS, syntax gates for Ruby/PHP, structural lints for SQL and Dockerfilesbeta.27
- 43-scenario calibration harness with CI floorsbeta.27
- Verification runs once per turn, roughly 2x fasterbeta.27
- QEL now grades the runs it used to skip: a thrashing loop that force-completes, or a loosely worded request, no longer slips through unverifiedbeta.30
- Regression contracts and Drift Radar: re-prove the contracts QEL already verified against your current code, report-only and air-gap-safebeta.31
- Proof-carrying commits: a signed, locally verifiable QEL trailer on the commit you verifiedbeta.31
- Post-loop code review: after QEL passes, an optional pass flags single-responsibility violations, weak names, and obvious bugs in the files the agent changed. Non-blocking and off by defaultbeta.31.3
- Rubric outcome grader: attach a free-text rubric and, once QEL passes, a one-shot grader pins a verdict, score, and justification to the verification cardbeta.31.3
Codebase Intelligence
- Codebase Wiki: a queryable knowledge layer over the dependency graphbeta.25
- Bodega Map: staleness tracking, project overview, and an agent-callable query_map toolbeta.26
- Managed llama.cpp embeddings for codebase searchbeta.26
Cost and Privacy Controls
- Spend caps: a hard ceiling on total cloud spendbeta.26
- Pre-send cost estimate in the status barbeta.28
- Air-Gap Active top-bar indicator, and air-gap extended to Git AI featuresbeta.26
- Local code review on the working diff: no PR, no cloudbeta.26
- Air-gap attestation: a signed, tamper-evident record of your network posturebeta.31
Editor and Vision
- Dockview panels: drag anything anywhere, split editorbeta.23
- Local two-model vision on llama.cppbeta.24
- Agent vision: screenshot and inspect the rendered previewbeta.21
- In-app docs hubbeta.26
- Hunk-level staging and a conventional-commit pickerbeta.27
- Dev-server-free preview: serve a plain static site or a generated HTML5 game (a bare index.html, no build step) straight into the Preview panel, on loopback onlybeta.31.6
In Progress
We ship small tagged releases frequently, so the honest in-progress list is short: it is whatever lands in the next tag. Here is what we have said out loud. While we are in open beta, everyone gets new features as they land, free.
Bodega One Code, in your terminal
- A native CLI with the same agent, local models, air-gap, and QEL verification, plus a headless mode for CIIn design. Free, no subscription. No date yet.
The Road to 1.0
- Hardening the beta into the full releaseNo date: it ships when it is ready. Pro, the $39 one-time commercial license, goes on sale then.
What comes next depends on what you tell us
Beyond the list above, we are not committing to a specific feature roster yet. What moves up in the queue is driven by beta feedback. If 50 users hit the same wall, that wall gets fixed before anything else.
We have a list of longer-horizon work - teams features, deeper agent capabilities, extended editor integrations. We will publish that list when there is something real to show, not as a promise board.
Join Discord to weigh in on what comes nextQuestions about
the roadmap
Is the beta open?+
Yes. The beta is free and open to everyone, commercial use included. Download free at bodegaone.ai/download. Full general availability comes later this year.
Will the roadmap change after launch?+
Yes. Priority shifts based on what beta users tell us. The in-progress items above are the ones we consider most important right now. That can change.
Do I get new features during the beta?+
Yes. While we are in open beta, everyone gets new features as they ship, free, with no early-access gate. Most weeks that means a new tagged release.
Can I suggest something for the roadmap?+
Yes. Join the Discord and post in the feedback channel. We read it. Enough signal on a feature moves it up. That is not a polite non-answer - it has already happened.
Where can I see the full changelog?+
Everything that has already shipped is documented in the changelog with version numbers and dates.
Shape what we build next.
Join the beta. Use the product. Tell us what matters. We read every piece of feedback and it directly influences what moves up in the queue.
Beta is live now, free for everyone. Full launch coming later this year. New features ship to all beta users as they land.