Opisense CodeTerminal and desktop
The same agent. At home in your code.
A coding agent that never compacts. Explore a codebase, plan the change and build it with the model you choose.
Then run opisense
Plan
Plan mode works out the approach before any code changes.
- Explore investigates, Plan proposes and Build makes the change
- Approve the plan, or start it as a goal
- Approved plans are saved as Markdown for later
❯Move auth from sessions to signed tokens with rotation.
◆ Explored 14 files
▸read src/middleware/auth.ts
Plan · 3 steps
1. Add a verifyToken helper in src/lib/token.ts
2. Add /auth/refresh with rotating refresh tokens
3. Replace the session check in authMiddleware
?Approve plan? [y] yes · [g] use as /goal · [e] edit
Memory
Observational Memory means the conversation never has to be compacted.
- Observations are extracted from every conversation
- Relevant context comes back in later requests
- Threads are saved per project and resume on their own
Thread resumed · billing-export
❯Pick up where we left off on the billing export.
◆ Recalled 3 observations
• The CSV export keeps the old column order
• Billing tests run with pnpm test:billing
• Totals are rounded per line, not per invoice
▸edit src/export/csv.ts
✓Continued without re-explaining the task
Goals
Goals keep it working across turns until the job is done.
- /goal sets an objective that can span many turns
- A judge model decides when it is done or needs you
- Skills and custom commands can run as goals
❯/goal Fix the failing release checks and open a PR when everything passes.
◆ Goal started · judge checks after each turn
▸bash pnpm test --filter release
- 3 tests failing in packages/release
▸edit packages/release/src/version.ts
▸bash pnpm test --filter release
+ All tests pass
✓Judge: goal complete · pull request opened
Subagents
Subagents take on parts of the work so the main thread stays focused.
- Built-in subagents to explore, plan and execute
- Each one keeps its own tool allowlist
- Pick a model for each with /subagents
❯Find why checkout is slow, then fix it.
▸agent:explore trace the checkout request path
Price lookups run once per cart line
▸agent:plan batch price lookups per cart
▸agent:execute apply the plan in src/checkout
+ const prices = await getPrices(cart.lines)
- for (const line of cart.lines) await getPrice(line)
✓Change applied and tests pass
Skills and plugins
Bring the tools and methods your team already uses.
- Install trusted plugins from a local folder or GitHub
- Connect MCP servers and Composio toolkits
- Hooks, skills and custom slash commands per project
❯/mcp
● linear 12 tools
● sentry 8 tools
● postgres 5 tools
❯/skills
release-notes · pr-triage · db-migration
❯/hooks
✓2 hooks loaded from .opisense/
❯Move auth from sessions to signed tokens with rotation.
◆ Explored 14 files
▸read src/middleware/auth.ts
Plan · 3 steps
1. Add a verifyToken helper in src/lib/token.ts
2. Add /auth/refresh with rotating refresh tokens
3. Replace the session check in authMiddleware
?Approve plan? [y] yes · [g] use as /goal · [e] edit
Try it in your terminal
One command to install. Or run it once with npx opisensecode.
Everything you need to ship
One agent for the whole loop: explore, plan, build and review.
Plan mode
Work out the approach before writing code.
Goals
Keep working toward an objective across turns.
Subagents
Hand parts of the work to focused agents.
Observational Memory
Context that carries over, without compaction.
Persistent threads
Conversations saved per project and resumed.
Skills
Turn a repeatable method into a skill.
Hooks
Run scripts around the agent’s tool calls.
MCP servers
Connect Linear, Sentry, Postgres and more.
Plugins
Add tools, commands and skills from trusted sources.
Any model
Claude, GPT, Gemini and more through one model router.
Your subscription
Sign in with Claude or ChatGPT.
Custom providers
Add OpenAI-compatible providers and models.
Pull request review
Review a GitHub pull request with /review.
Diffs
See modified files and the git diff with /diff.
Sandboxed scripts
Run Python and Node in an isolated VM.
Scheduled tasks
Run a prompt on a schedule.
Tool approvals
Approve each tool call, or opt in to auto-approve.
Usage tracking
Token counts per thread with /cost.
Quick answers
Bring Opisense to your terminal.
Install it now and start with your own codebase.