Objective agent
Objective is the coding and comms agent inside Objectiv. Objectiv is the shell (editor, terminal, marketplace). This guide focuses on how the agent behaves in the desktop app.
Shared harness
Objective reuses the A-Zing coding-agent tool surface: repository search, reading files, proposing unified diffs, running terminal commands, and recording trajectories with accept/reject labels. That alignment means:
- Playbooks you test in Objectiv can match CI batch jobs.
- Pilot users can compare IDE sessions with logged trajectories.
- Future VS Code extension surfaces can share the same tool definitions.
Coding tools (Community + Enterprise)
Available without a paid seat:
- Plan and execute refactors across the active workspace root
- Run tests and linters via the integrated terminal
- Iterate on failures until you approve a patch
- Cancel mid-turn when the model diverges (streaming)
Recommended setting: objectiv.agent.approveBeforeApply: true so disk writes always follow human review.
Approve-before-apply workflow
- Agent proposes a patch (one or more files).
- UI shows a diff summary—you accept, reject, or ask for changes.
- On accept, files write to the workspace; on reject, nothing changes.
- Follow-up turns can depend on terminal output you trigger manually or via the agent.
This mirrors how responsible teams want comms to work too—draft first, send after approval (Enterprise).
Comms tools (Enterprise only)
When a valid Enterprise JWT is present, additional tools may draft and send via configured providers:
- Chat — Teams, Slack, Mattermost, Google Chat
- Mail — IMAP/POP mailboxes you configure
- Meetings — Zoom, Teams, Slack calls, Google Meet
Community keeps Slack browse/history with your token but not agent send. Sidecar enforcement rejects unlicensed comms calls in production builds. See Enterprise guide.
Active root & context
In multi-root workspaces, Objective targets the active folder for edits and terminal cwd unless you specify otherwise in the prompt. Keep unrelated repos in separate roots to reduce cross-repo mistakes.
Practical tips
- Scope tasks (“only under
packages/api”) for faster, safer turns. - Ask the agent to run tests before offering a patch—you still approve the write.
- Prefer small follow-up turns over one giant “rewrite the app” prompt.
- For regulated environments, pair local Ollama with approve gate—see models guide.