Agents Workspace

Get more done while your agents work.

Run several AI agents at the same time. Odina keeps every task organized and tells you when something needs your attention.

Built for Codex, Claude Code, Cursor, Kimi, and any CLI agent.

Live workspaceFour jobs in motion
4 active
01
CodexFix checkout retries
Working
02
ClaudeDraft launch brief
Done
03
CursorMigrate auth callback
Needs you
04
KimiAudit route performance
Working
Open

One workspace for every agent

Launch Codex, Claude Code, a shell, or any CLI agent from the project folder where it needs to work.

01 / 04

Inside Odina

Odina
1 agent working
ORCHESTRATORCheckout retry tests
1 working0 need you0 ready
Checkout retry testsodina-web

● mcp__github__get_pull_request(odina/web, #418)

└ Loaded 7 changed files and 3 review comments

● Search(src/billing/**, "retryDelay")

└ Found 6 matches in 4 files

• The checkout hook retries linearly, while the recovery worker expects exponential backoff. I’m aligning both paths and adding the timeout case.

● Read(src/billing/retry.ts, tests/checkout/retry.test.ts)

└ Read 284 lines (ctrl+r to expand)

● Update(src/billing/retry.ts)

└ Updated 2 files with 19 additions and 6 removals

- const delay = attempt * 1000

+ const delay = Math.min(baseDelay * 2 ** attempt, MAX_DELAY)

● Bash(bun test tests/checkout/retry.test.ts)

◌ Running 24 checkout tests… 18 passed · 6 remaining

auto-accept edits onshift+tab to cycle
Launch brief & PDFodina-docs

● mcp__notion__fetch(Launch plan)

⎿ Read roadmap, positioning, and 14 launch tasks

● mcp__slack__search("customer proof")

⎿ Found 23 messages across #feedback and #sales

● Write(docs/launch-brief.md)

⎿ Wrote 1,842 words with linked source notes

● Write(docs/launch-checklist.docx)

⎿ Created 6 sections, 18 owners, and 32 tasks

● Bash(render-pdf docs/launch-brief.md)

⎿ Rendered and visually checked 8 pages

✓ launch-brief.pdf and launch-checklist.docx are ready

auto-accept edits onshift+tab to cycle
Auth callback migrationodina-api

● mcp__linear__get_issue(AUTH-284)

└ Loaded acceptance criteria and linked incident

● Read(src/auth/callback.ts, prisma/schema.prisma)

└ Traced PKCE verifier lifecycle across 5 modules

● Update(src/auth/pkce.ts)

└ Added verifier encryption and callback validation

● Bash(bun test auth)

└ 18 passed · 0 failed · 3 snapshots updated

Permission required: bun prisma migrate deploy

auto-accept edits onshift+tab to cycle
Route performance auditodina-api

◉ mcp__sentry__search_issues · /api/usage latency

✓ 1,284 traces · p95 1.82s · 61% duplicate queries

◉ Explore · mapped 12 route handlers and 4 middleware paths

◉ ReadFile · src/api/usage/route.ts

✓ Found duplicate account and entitlement lookups

◉ StrReplaceFile · route.ts

Cached account lookup for the request lifecycle

◉ Bash · bun run bench:api

◌ Benchmarking 500 requests… p95 684ms ↓62%

auto-accept edits onshift+tab to cycle
Tell Odina what you want to get done...
Coming next
Why Odina

Agents scale.
Human attention doesn't.

Starting another agent is easy. Following every terminal, question, and result is the hard part. Odina makes parallel work manageable.

01

Start the whole queue

Give separate jobs to Codex, Claude Code, or any CLI agent and put them all in motion instead of waiting on one task at a time.

02

See everything moving

Keep every agent, repository, and task visible in one workspace. Jump into any terminal without losing the rest of the picture.

03

Step in only when needed

Keep doing your own work. Odina brings you back for completions, blockers, questions, and approvals—with enough context to act.

Privacy & security

Your work stays
on your machine.

Odina runs terminals and agents locally. Your source files are not uploaded to an Odina account, and your terminal history and project memory remain on your Mac.

01

Local by default

Agents, shells, files, terminal transcripts, and project memory live on your device—not in an Odina cloud workspace.

02

AI only when it helps

Optional summaries, smart copy, memory distillation, planning, and voice send only the context needed for that request to the configured AI provider. Supported text requests redact likely secrets first.

03

No Odina data warehouse

Odina does not retain source code, terminal transcripts, project memory, or raw voice recordings in an account database. Provider security and abuse-monitoring policies may still apply.

Odina currently uses the OpenAI API for optional hosted AI features. OpenAI states that API inputs and outputs are not used for model training by default. Review OpenAI's business data policy ↗

Coming next

From managing terminals
to orchestrating outcomes.

Soon, tell Odina the outcome through chat or voice. It will open the right folders, start the right agents, coordinate the work, and return when your judgment is needed.

Your intentOdina orchestratesAgents executeResults return

The operating layer between humans and AI agents.

Run more agents.
Stay in control.

Download Odina