Unify the nav bar
Nav implementation
Plan
BottegaFeatures / Agents
Switch Agents, keep your workspace.
Run Codex, Claude Code, Kimi Code, and OpenCode in the same local workspace—using their official CLIs and the subscriptions you already have, with no extra fees.
01 · Multi-Agent
Official CLIs. One place to work.
Bottega runs Codex, Claude Code, Kimi Code, and OpenCode through their official local CLIs. It does not replace their harness with a generic Agent layer.
Authentication, subscription access, and quotas stay with the provider. Model and reasoning-effort controls come from each CLI's live catalog, so an option appears only when that backend can actually accept it.
A conversation stays bound to the Agent that started it. Open another conversation when you want a different Agent; both can live in the same Project.
Ship the release notes
Which PRs merged since 0.1.9 would a user actually notice?
Six of the eighteen. Agent parity is the one people will feel first; the two internal refactors fold into a line.
Draft the 0.2.0 release notes from the merged PRs since 0.1.9.
Plan
0.2.0 release notes
Summary
- Six headline changes since
0.1.9, grouped by what a user notices first. - Agent parity is the lede —
KimiandOpenCodenow reach the same permission ladder asCodex. - Two internal refactors fold into one line; nobody outside the repo felt them.
Sequence
- Group the 18 merged PRs by the surface a user touches, not by the package they landed in.
- Write each entry as what changed — the PR number is provenance, not news.
- Mirror the file into
docs/, then read both back so the two can never disagree.
Files
CHANGELOG.md— six entries, newest first, absolute dates.docs/changelog/README.md— the public mirror, same six entries.
Not in this pass
- The
0.1.xbackfill — older entries are a separate decision about how far back the record goes. - Release tagging and the download page; they read this file, they do not write it.
Plan above. Say go and I will write it; say which grouping you want changed and I will redraw it first.
Responding
Ask anything
Approve for me
02 · Conversation parity
One interaction model, not one Agent.
Each Agent Harness has its own runtime model. To keep the experience consistent, Bottega provides one interaction layer and adapts streaming replies, tool status, Plan mode, queued messages, and more to each backend.
Here is the current capability coverage, which continues to evolve:
CapabilityCodexClaudeKimiOpenCode
Streaming replies✅✅✅✅
Model / Thinking selection✅✅✅✅
Tool status✅✅✅✅
Plan mode✅✅✅✅
Queue mode✅✅✅✅
User approvals✅✅✅✅
@ mentions✅✅✅✅
Browser Use✅✅✅✅
Fork✅✅✅✅
Edit previous message✅✅✅✅
03 · Cross-Agent collaboration
Beyond a unified experience, Agents finish the work together.
Let Claude frame the problem and create a Plan, then pass the necessary context to Codex for implementation. When it is done, the result returns automatically to the original Chat for review, revision, or handoff to the next Agent.
Each Chat keeps its own Agent, workspace, and task history. Messages can run immediately or wait in a queue; the process stays visible, and every result remains traceable and reusable.
Unify the nav bar
Nav implementation
[From Section @Unify the nav bar (source_section_id=sec_7f2a3c)]
@Unify the nav bar · also verify the mobile breakpoints
Unify the nav bar
Nav implementation
[From Section @Nav implementation (source_section_id=sec_91b4e0)]
Plan intent and implementation match; no blocking issues.
First let Claude turn the problem
into a Plan
Drop the work into
another Chat's queue
When it finishes, the result returns on its own
Back in the original Chat — review it,
or hand it to the next Agent