OpenAna named a Gartner Cool Vendor 2025

Ana SDE · VS Code & IntelliJ

The engineerin your IDE.

Chat with Ana about the code you have open, or let the agent plan, build, run and debug - in five modes, following your team’s rules, using your tools through MCP, and asking before every step.

Visual Studio Code logoVisual Studio CodeIntelliJ IDEA logoIntelliJ IDEAModel Context Protocol (MCP) logoModel Context ProtocolGet Ana SDE

Ana SDE · in your IDE

Your editor,
with an engineer in it.

Ana SDE brings the Software Engineer into VS Code and IntelliJ. Chat about the code you have open, or let the agent plan, edit, run and debug - and approve every step before it happens.

  • Five modes

    Architect, Code, Ask, Debug and Orchestrator - each with the right tools for the job.

  • Agent, not autocomplete

    Edits files, runs commands and tests web apps across the workspace.

  • Approval at every step

    Nothing changes and nothing runs without your yes.

  • Hands off to the platform

    Bigger work continues as a tested pull request from Ana’s Software Engineer.

Ana for VS CodeAna for IntelliJModes, rules & MCP →
client.ts - paymentsVS Code · IntelliJ
1export async function charge(req: ChargeRequest) {
2  return withRetry(() => api.post('/charges', req), {
3    attempts: 3, backoffMs: 250,
4    idempotencyKey: req.requestId,
5  });
6}
ArchitectCodeDebugAskOrchestrator

Plan retries for the payments client

AnaPlan: wrap charge() in exponential backoff, cap at 3 attempts, make it idempotent with the existing request id. 3 files.

Five modes

The right tools
for each kind of work.

Modes aren’t just prompts - each one has its own tool set and file-access rules, enforced in code.

Architect mode

Plan before anyone types.

What it does

Reads the codebase, designs the change, writes the plan and design notes.

Boundaries

Can edit Markdown files only (plans, design notes) - never source code.

Rules

Your conventions, in every change.

Commit a .ana_instructs file at the root of the repo and Ana follows it on every task in that workspace - style, architecture boundaries, test requirements, the commands to run before a PR. Add personal custom instructions on top.

  • Versioned with the code, reviewed like code
  • Same rules for everyone on the team
  • Per-mode behaviour on top of project rules
.ana_instructs
# .ana_instructs - committed at the repo root, shared by the whole team

- TypeScript strict mode. No `any` without a comment explaining why.
- Every change to src/payments/** needs a unit test and an idempotency check.
- Use our logger (src/lib/log.ts), never console.log.
- Database access only through repositories in src/db/repos.
- Never edit generated files under src/gen/** - change the schema instead.
- Before opening a PR: npm run lint && npm test.
MCP settings · mcpServers
{
  "mcpServers": {
    "internal-docs": {
      "command": "node",
      "args": ["./tools/docs-mcp/server.js"]
    },
    "postgres-readonly": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://readonly@db/app"]
    },
    "jira": {
      "command": "uvx",
      "args": ["your-jira-mcp-server"],
      "env": { "JIRA_TOKEN": "…" }
    }
  }
}

MCP

Bring any tool with MCP.

Ana SDE is a Model Context Protocol client. Point it at MCP servers - community ones or your own wrapping internal APIs, databases and docs - and Ana can call their tools and read their resources while it works.

  1. 1DeclareEach server is a command, args and env in the MCP settings file.
  2. 2Live reloadEdit the file and Ana reconnects - no restart.
  3. 3ApproveEvery tool call and resource read is shown first, one at a time.
MCP integration details →

Workflows

How teams chain the modes.

Ticket to PR

  1. Ask: understand the area
  2. Architect: plan the change
  3. Code: implement + tests
  4. Debug: fix what fails
  5. Hand off as a pull request

Bug from production

  1. Debug: reproduce from logs (via MCP)
  2. Ask: find sibling call sites
  3. Code: fix + regression test
  4. Checkpoint, verify, PR

Large refactor

  1. Orchestrator: split into subtasks
  2. Architect: design per module
  3. Code: implement module by module
  4. Debug: run the suite between steps

Safety in the editor

Powerful, and
never unsupervised.

The same guardrails philosophy as the rest of Ana - applied at the keystroke level.

  • ✓

    Approve every step

    File edits, terminal commands, browser actions and MCP calls are shown before they run. Nothing happens without your yes.

  • ✓

    Mode boundaries

    Each mode has a curated tool set and file-access rules enforced in code - read-only modes cannot write.

  • ✓

    Checkpoints

    Ana snapshots your workspace as it works, so you can compare or roll back to any earlier point.

  • ✓

    Your rules, enforced

    Project rules in .ana_instructs and your own custom instructions shape every plan and change.

  • ✓

    One MCP call at a time

    Ana waits for each MCP result before the next call - no fan-out you didn’t see.

  • ✓

    Org-controlled access

    IDE access is tied to your organization’s Ana account and roles.

Safety & privacy at OpenAna →Guardrails →

FAQ

Ana SDE, answered.

Visual Studio Code and IntelliJ-based IDEs.

Get Ana SDE

Open your editor. Meet Ana.

VS Code or IntelliJ, your repo, your rules - and approval on every step.

Get access
Software Engineer