Opifer 1.6.0 — web search for everyone, archived conversations

Added

• Web search without setup: the models of ChatGPT (subscription) and Anthropic search by themselves, with their provider's own tool, on every turn. For other models, Settings → Web search stores a Brave or Tavily key, or a SearXNG URL, as a company secret (PUT/GET/DELETE /v1/companies/:id/web-search); the server's environment keys still work for every company. web_search is offered only where a source exists.
• Archive a conversation from the Chat list (and bring it back): POST /v1/sessions/:id/reopen.

Opifer 1.5.0 — files for a task

Added

• Files for a task: Give a file in the task panel and o4r task upload <task> <file...> write into the task folder (uploads/ by default) and leave a comment for the agent; PUT /v1/tasks/:id/files/<path> with the raw bytes.

Opifer 1.4.0 — email and a browser

Added

• Email as a connection: a mailbox per company (SMTP to send, IMAP to read), the password a company secret. The agents get <name>__send (high risk: a person approves each email unless the policy allows it), <name>__list, <name>__read and <name>__search. Connections → Tools → Add a mailbox; o4r connection add-email <name> --smtp host:port --imap host:port --user … --from …. Migration 0012_email_connections.

• A browser for the agents: with Chrome or Chromium on the machine (found at start, OPIFER_BROWSER, or browser.executablePath in config.json), browser_open, browser_read, browser_click, browser_type, browser_select and browser_screenshot (a PNG in the task folder). One page per session, public addresses only, closed after ten minutes idle; o4r doctor reports the browser in use.

Fixed

• The CLI declared a JSON body on calls that send none, so o4r connection check|remove, o4r resume, o4r routine run, o4r skill restore and the chat interrupt answered Bad Request in 1.3.0.

Opifer 1.3.0 — software work and artifacts

Added

• Projects can be git repositories (repoUrl, branch): cloned into the project's folder at creation, with the company's GITHUB_TOKEN for private ones; agents push with a token bound to them through a credential helper, never stored in the repository. o4r project list|create --repo --branch; repository URL and branch on the project form, clone status on the project.
• Tools for software: edit_file (exact replacement), apply_patch (unified diff), and run_coder, which hands a brief to Claude Code or the Codex CLI installed on the machine and reports the summary and the diff stat (found on the PATH at start, or coder in config.json; high risk). Tasks on a repository carry an engineer's guide; the terminal sets git up and can run up to 30 minutes.
• Artifacts: everything the agents produced, in the task panel (products declared at delivery, files of the task folder, open or download) and in Work → Artifacts for the whole company; GET /v1/companies/:id/artifacts, GET /v1/tasks/:id/files[/path]; o4r artifacts, o4r task files <id> [path].
• Archive an agent from the Team page (Archive / Bring back, an Archived list) and o4r agent list|pause|resume|archive|restore.

Changed

• The default sandbox image is node:22-bookworm (git, curl, python, build tools) with the network on; "sandbox": { "network": "none" } isolates it.
• No made-up default model: without a connected provider the interface, the CLI and the runtime say so.
• Migration 0011_project_repos.

Opifer 1.2.0 — o4r update

Install or update

npm install -g @opifer/cli      # first time
o4r update                      # from now on

Added

• o4r update: installs the latest @opifer/cli from npm where this one is installed and restarts the server (under systemd or launchd the service comes back by itself); --check only says whether a newer version exists; from a repository checkout it prints the git and pnpm commands.
• The server looks at npm once a day for a newer version (updates.check: false in config.json turns it off; nothing else leaves the machine); the health check, o4r doctor and Settings say when one is out.

Fixed

• The interface is English by default whatever the browser language; Italian is chosen in Settings (1.1.1 of the interface).
• The hidden password prompt of o4r user password and o4r auth enable failed in a terminal (1.1.2 of the CLI); o4r doctor crashed on the companies list in authenticated mode (1.1.1 of the CLI).

Full changelog: https://github.com/NextEpochs/opifer/blob/main/CHANGELOG.md · Website: https://opifer.dev

Opifer 1.1.0 — authenticated mode

Install

npm install -g @opifer/cli
o4r init --company "My company" && o4r up            # your machine
o4r init --company "My company" --auth --email you@example.com   # a server: sign-in required

Opifer can now run on a server or a VPS: o4r init --auth --email you@example.com (or o4r auth enable) and every call to the interface and the API needs a signed-in person or an API key.

Added

• People with email and password (scrypt), session cookie (HttpOnly, SameSite=Lax, Secure behind HTTPS, 30 days sliding), ten failed sign-ins a minute per address.
• API keys (opk_…) for the CLI and integrations, shown once, stored hashed, revocable.
• Roles, server-wide: observer reads, operator does the daily work, admin configures, owner manages people, keys, the emergency stop and export/import. The last owner cannot be removed.
• CLI: o4r auth enable|disable|status, o4r user add|list|remove|password|role, o4r apikey create|list|revoke; o4r init --auth --email --password; the CLI signs its own calls with a key in <home>/credentials/cli.key.
• Interface: sign-in screen, the signed-in person in the sidebar, sign-out in Settings; preferences per person.
• The interface works under a path (https://example.com/opifer/): relative assets, API, events socket and service worker.
• Docker: OPIFER_AUTH_EMAIL and OPIFER_AUTH_PASSWORD turn authenticated mode on at first start.
• Migration 0010_auth: users.password_hash|role|status|last_login_at, user_sessions, api_keys.

Changed

• o4r doctor reports the authentication mode instead of warning about exposure when sign-in is on.
• The guide has a section "On a server" with the nginx snippet; the security review describes both modes.

Full changelog: https://github.com/NextEpochs/opifer/blob/main/CHANGELOG.md · Website: https://opifer.dev

Opifer 1.0.0

Install

npm install -g @opifer/cli
o4r init --company "My company" && o4r up   # http://127.0.0.1:4700

Node.js 22. PostgreSQL is embedded, Docker is optional. Website and guides: https://opifer.dev

The repository is public, the website is up at opifer.dev, and everything the README says has been verified on a second machine. No breaking change from 0.1.0: the version says the product is ready to be tried.

Added

• Website site/ (home, the guides rendered from docs/), built with node site/build.mjs.
• CONTRIBUTING.md, CLA.md (contributor licence agreement, sign-off on every commit checked by a workflow), SECURITY.md with private vulnerability reporting, issue forms and a pull request template.
• The task form creates a project inline.
• DockerEnvironment option user: on Linux the sandbox runs as the server's user, so files written under the task folder belong to it on the host.

Fixed

• The container image did not build (pnpm prune --prod without a TTY) and then did not start (commander, then @opifer/db missing from a production install); verified on Linux with Docker 29: start, demo, restart, recovery after a kill.
• docker-compose.yml published the port on every interface; it binds 127.0.0.1 now.
• o4r up refused to start after a container restart because the stale pid file pointed at pid 1, its own pid.
• After a server restart every wake-up left running went back to the queue without comparing the database clock with the process clock (a random CI failure).
• GET /v1/sessions/:id with a malformed id answered 500; it answers 400.
• The CLI printed 0.0.0.0 in URLs when listening on every interface.

Changed

• The demo company is called Proclive.

Full changelog: https://github.com/NextEpochs/opifer/blob/main/CHANGELOG.md · Website: https://opifer.dev

Opifer 0.1.0 — the MVP

Changelog

All notable changes to Opifer. The format follows Keep a Changelog; versions follow SemVer.

0.1.0 — 2026-09-20 — the MVP

The first release: the scene in section 16.1 of the specification runs end to end, in tests and live.

Foundations (M0)

• pnpm/TypeScript strict monorepo, embedded PostgreSQL, forward and backward SQL migrations mirrored by a Drizzle schema and a test, immutable audit log, company_id on every domain table, /v1 API, o4r CLI, React interface, CI on Linux and macOS, third-party notices generated.

Runtime (M1)

• Phased agent loop with a stable prompt prefix, retries and fallback model, no tool replay after a restart, native tools (terminal, files, search, ask), operator injection mid-turn.
• Providers: Anthropic, OpenAI (API key and ChatGPT subscription sign-in), OpenAI-compatible endpoints for local models.

Governance (M2)

• Budget reserved before every model call, caps per company/agent/project/task/turn with monthly/daily/lifetime windows; approvals; permissions per role and agent; dangerous commands always ask; encrypted, versioned secrets injected only into tool calls and redacted from output; every agent change is a revision.

Work (M3)

• Goals, projects, tasks with atomic checkout and leases; delegation downward; delivery for review; done means verified; comments and mentions; wake-ups as the at-most-once queue; the scheduler.
• Interface: Home board of widgets, Inbox, Team as an org chart with drag-and-drop hiring, Work board, Chat, Money, Settings; English and Italian; dark and light.

Learning (M4)

• Memories and skills at agent, team and company scope; snapshot in the prompt; background review on a copy of the conversation; curator; promotion under a company policy; a repeated job costs 45% less in the test suite.

Connections and routines (M5)

• Routines (interval, cron, once) at most once per due time, in a session or as a task the agent delegates and gets reviewed; delegation closes the loop (the delegator reviews, approves or sends back).
• MCP servers and workflow endpoints as governed tools; inbound webhooks with hashed tokens; outbound signed events with retries; Telegram bot with pairing codes and one-tap approvals; Docker sandbox with no network when available.

The complete MVP interface and robustness (M6)

• Emergency stop; two-line context compression, traced; agents can act in a conversation (company status, hand out work).
• Preferences on the server; plurals; money by language; installable app (PWA); WCAG 2.1 AA with zero axe violations on every page; Italian parity test.

Finishing (M7)

• Company export and import; demo company; o4r doctor extended; quickstart and security review; load test (20 agents, 200 tasks, 10 concurrent runs on 2 vCPU: 220 runs in 13 s, none failed); scene 16.1 as a contract test.