Documentation Index
Fetch the complete documentation index at: https://docs.replicas.dev/llms.txt
Use this file to discover all available pages before exploring further.
CLI. The Replicas CLI now supports
replicas env for listing, creating, editing, and deleting environments, plus managing environment variables and files.Replica API. API keys are now available to Hobby organizations from Settings → API Keys.Bug Fix. Claude reasoning blocks now display correctly in the workspace chat and Linear session updates.Domain migration. Our primary domain is now
tryreplicas.com. All URLs (dashboard, API at api.tryreplicas.com, docs at docs.tryreplicas.com, and CLI install script) now use the new domain. The previous replicas.dev domain will redirect automatically, and email addresses remain on @replicas.dev.UI Polish. The Warm Hooks tab in Settings → Environments now shows the full execution chain (Global hook above, per-repo replicas.{json,yaml,yml} hooks below) inline with the environment-level editor.UI Polish. Warm pool failure logs in Settings → Environments are now collapsible instead of displayed inline, and preserve line breaks for readability.UI Polish. The Warm Hooks editor now has separate Save and Test buttons so you can persist changes without waiting for a test run, or test scripts without committing them.CLI. Added preview delete and preview remove commands for cleaning up preview URLs from workspaces.Bug Fix. New dashboard workspaces now respect the configured default agent instead of always starting with Claude.Bug Fix. Fixed an issue where testing warm hooks that take over 60 seconds would timeout with no output visible. Save & Test now streams progress via SSE so the connection stays alive and results are always returned regardless of execution time.Engine. Hook commands (warmHook and startHook) now support a separate option that controls whether commands run in isolated shell processes (the default) or as a single script sharing shell state. Set separate: false to run all commands in one shell session, sharing environment variables and working directory changes.UI Polish. Workspace Media tiles now open in a navigable lightbox — use the chevrons to move between items.UI Polish. All media surfaces (tiles, lightbox, inline images, video and audio players) now have a download button.
Slack DMs. You can now message the Replicas bot directly in a 1:1 or group DM without an
@Replicas mention, and replies in the bot-created thread continue routing to the same workspace just like in channels. Existing Slack installs need to re-authorize from Integrations so the new im:* and mpim:* scopes are granted before DMs deliver. Read moreDeep-linkable settings. Settings sections (Automations, API Keys, Agents, GitHub, Integrations, Preferences, Billing, Members, Audit Logs, Environment) each have their own URL now (e.g. /dashboard/billing), so you can bookmark and share them, and browser back/forward navigates between sections. Old /dashboard?section=… links continue to redirect.Bug Fix. Videos uploaded to the workspace Media tab now load correctly with native browser playback controls.Codex MCP support. Stdio and streamable
http MCPs configured under Settings → Environments → MCPs are now available inside Codex sessions. sse MCPs remain Claude-only since Codex does not support that transport.Forward-deployed engineer on Team and Enterprise. Team and Enterprise plans now include a dedicated forward-deployed engineer who runs a 60-minute onboarding call to learn your stack and codebase, then sets up your environments for or alongside you. Read the docs!UI Polish. Selecting sse as the transport in the MCP form now shows an inline warning that the entry will be skipped for Codex agents.Bug Fix. PR comments and reviews now follow the same gate as issue comments — Replicas no longer replies to non-@tryreplicas mentions outside the GitHub Auto Responders allowlist.Bug Fix. Fixed an issue where the workspace Media tab rendered every uploaded image or video as an empty tile.Bug Fix. Toggling a warm pool from Settings → Environments → Warm Hooks & Pool now shows a success or error toast.Bug Fix. Fixed an issue where enabling a warm pool on an environment in a Hobby or Developer org did not actually pre-warm any workspaces.Replica API.
POST /v1/replica accepts repository_set_id or repository_ids again as a backwards-compatible alternative to environment_id. The server resolves either selector to the first matching environment (preferring the auto-created Default {repo} env). The legacy fields are marked deprecated; please migrate to environment_id when convenient.Bug Fix. The “Claude Code is thinking…” indicator no longer hangs after a turn ends when the last action started a long-running background task.
UI Polish. Account Settings → Preferences now shows a warning pill on the Attribute PRs to me setting when GitHub is not connected via OAuth. Clicking the pill jumps straight to the Connections tab so you can finish wiring it up.
Bug Fix. Toggling a setting in Account Settings → Preferences no longer shifts surrounding content while saving.
Bug Fix. Audit log rows and the user filter now display the actor’s email or name (falling back to
API key <name> or Unknown User) instead of a truncated user ID.May 3, 2026
Default Environment as a single env id, with Slack and Linear intake picking an environment
Default Environment stored as a single env id. The org-level Default Environment setting is now stored as a single environment id (instead of a tagged union over repo, repo set, and env). Existing settings are migrated automatically: a default repository or repository set is replaced with the environment bound to it. The dashboard, Slack intake, and Linear intake all read this value directly with no resolution step.Environment selection in Slack and Linear intake. Slack
@Replicas mentions and Linear agent sessions now ask “Which environment should I use?” instead of picking a repository or repository set, so multiple environments bound to the same repo are independently reachable. The Slack inline tag is now [env:name] (replacing [repo:name] / [set:name]). Read more · LinearDefault Environment in Org Settings. The org-level Default Workspace Target setting is now Default Environment, populated from the org’s environments.Environments as the primary primitive. Environments replace the four-way scope hierarchy (Global, Repository, Repository Set, Custom). Each org has one Global environment plus any number of named environments; non-global envs optionally bind a repository or repository set as a property. Variables, files, skills, and MCPs all belong to a single environment, and a workspace’s environment is merged on top of Global. Each environment also gets its own warm hooks, warm pool, and optional system prompt injected into the agent. Read the docs!Attach an environment to an automation. Automations now run in exactly one environment, picked from the Environment dropdown in the automation editor. The environment supplies the repo and the resolved vars / MCPs / skills. Read moreBug Fix. Removed an erroneous upgrade paywall on the Automations page for Hobby orgs (Automations have been available on Hobby, capped at 2, since April 26).Bug Fix. Removed an erroneous upgrade paywall on the Warm Hooks & Pools tab for Hobby and Developer orgs — both features are available on all plans.Bug Fix. External messages (GitHub CI failures,
@tryreplicas comments, Slack/Linear replies) now go to the workspace’s most recently used agent instead of its default.Bug Fix. The dashboard sidebar now remembers its open/closed state across reloads, only auto-collapsing on narrow viewports.UI Polish. Workspace chat messages now show a clock time in the header (full date/time on hover), and each completed agent turn renders a Replied in 1m 23s marker below it.Codex Quota Status. When a Codex workspace runs out of OpenAI credits or hits a rate limit, the agent chat now shows an amber pill explaining the state instead of looking stuck.Audit Log. A new Audit Log section under Organization settings records every configuration change in your org — automations, environments, connectors, members, org settings, and API keys — with the actor, action, resource, timestamp, and a redacted before/after diff. Filter by resource type, action, or actor, search across rows, and expand any entry to inspect the diff. Read the docs!
Bug Fix. Brand-new workspaces now show a “Provisioning workspace…” indicator instead of a stuck “Claude Code is thinking…” timer until the sandbox is ready.Bug Fix. Typed drafts in the workspace chat are no longer lost when the workspace finishes provisioning.
Bug Fix. Workspace chat drafts are now preserved per-workspace across tab switches and page reloads.
Plan Quote-Reply. The dashboard plan viewer now supports Google-Docs-style quote-reply: select any text in a plan to get a floating Reply button, and your replies collect in a Comments rail alongside the plan. Hit Send to ship every quote and reply in a single message — they render as structured blocks in the chat so the agent knows exactly what each reply is responding to. Read the docs!
Confirm before deleting environment variables. Deleting a variable from Settings → Environments → Variables now opens a confirmation dialog showing the variable key, so a single misclick on the trash icon no longer removes a variable.Bug Fix. Toggling an automation’s enabled switch now updates only that row instead of flashing the entire Automations table.
Bug Fix. A freshly-created workspace no longer briefly shows the “Claude Code is thinking…” indicator before any prompt has been sent.
Pricing update. Developer is now $120/seat/month (was $30) and Team is now $300/seat/month (was $120). Read the docs!Hobby gets Automations and warm pools. Both features are now available on every plan, with rate limits scaling by tier. Hobby is capped at 2 automations, 5 executions/hr/org, 2 executions/hr/automation, and a 60-minute minimum cron interval. Read moreHobby split usage buckets. New Hobby orgs now receive two separate 20-hour buckets at signup (one-time, not monthly). One covers human-initiated workspaces (Dashboard, Slack, Linear, GitHub), and the other covers Automation usage, so automations can no longer drain dashboard hours.Included automation minutes. Developer includes 5,000 automation/API minutes per billing cycle and Team includes 15,000, before $0.0166/min metered overage applies.Repository limits. Hobby is capped at 3 repositories, Developer at 10, Team and Enterprise unlimited. Repository-set membership counts toward the cap (a repo in two repository sets consumes three slots).Auto-upgraded sandbox resources for Team. Team workspaces are auto-bumped to 32 GB disk and 16 GB memory.Bug Fix. The row icon in the Automations table now matches the automation’s most frequent trigger type (Slack, Sentry, custom webhook, etc.), not always GitHub.
Doppler Integration. Connect Doppler to sync environment variables from your Doppler configs into workspaces. Configure a Service Token per environment scope (Global, Repository, or Repository Set) from the Integrations section in the Variables tab. Secrets are automatically fetched at workspace creation time, and manual variables still take priority. Read the docs!Draft PRs by default. A new Open new PRs as drafts toggle in Account Settings → Preferences tells workspace agents to open every new pull request as a draft, with the same user to org cascade as PR attribution. Off by default, and you can still ask the agent to open a non-draft PR for a specific task. Read the docs!Rebase / Merge Main From Sidebar. Workspace cards in the sidebar now have a branch icon that opens a popover with Rebase on main and Merge in main actions. Selecting one sends the corresponding message to the workspace’s active agent chat.Deep-Linkable Account Settings. The account settings modal now mirrors its open state into the URL via an
accountSettings query param (connections, agents, preferences), so reloading or sharing a link reopens the modal on the right tab. The “connect GitHub” error shown when an org enforces the GitHub-required security policy now includes a direct link to dashboard?accountSettings=connections in Slack, Linear, and API responses.Bug Fix. Diff viewer file trees now show the correct added and removed line counts (previously wrong on files with only additions or only deletions).April 23, 2026
Sentry integration, custom webhook automation triggers, and MCP servers in Environments
Sentry Integration. You can now connect Sentry from Organization Settings so Replicas can react to Sentry events. Admins approve the install from Sentry once, and the connection is ready to use in automations. Read the docs!Sentry Automation Trigger. Automations can now be triggered by Sentry issue alerts, new issues, and new error events, with optional filters for Sentry project and minimum severity level. Read moreCustom Webhook Automation Triggers. Automations now support a custom webhook trigger with a unique URL per automation.
POST any JSON body to fire the automation, and the payload is embedded into the prompt under ## Trigger Payload. Useful for wiring up systems we do not have a first-class integration for. Read moreMCP Servers in Environments. Configure MCP servers at the global, repository, or repository-set scope from Settings → Environments → MCPs. Supports stdio, http, and sse transports, with headers and env vars stored encrypted at rest. Scopes are unioned at workspace provisioning, matching the Skills inheritance model. Read the docs!Smarter Linear Context. Agents triggered from a Linear issue now receive prior comments on the issue as context, not just the description and the triggering comment. This gives the agent the same background a teammate would have when picking up the work.Bug Fix. ANTHROPIC_API_KEY and OPENAI_API_KEY env vars no longer silently override the workspace’s configured auth (Claude OAuth, Bedrock, or Codex OAuth).Bug Fix. Workspaces running an older engine no longer fail to report status and PR URLs back to the dashboard and Linear.Automations GA. Automations are out of beta and available to all qualifying plans. Read the docs!Codex SDK Update with GPT-5.5 Support. Updated the @openai/codex-sdk from 0.111.0 to 0.124.0 and added GPT-5.5, GPT-5.4 Mini, and GPT-5.3 Codex Spark to the Codex model picker. Workspaces using Codex will now use GPT-5.5 by default.Bug Fix. The
install.sh installer now updates the appropriate shell rc file so the replicas command lands on PATH in new terminals, including for zsh and Homebrew-Bun setups.Multiple PRs per repo in workspace view. A workspace that opens more than one pull request in the same repository now shows every PR in the agent chat’s “View PRs” popover and mobile dropdown, labeled by PR number. Previously only the most recent PR per repo was surfaced.Bug Fix. Warm and start hook logs with large output now show a preview and a “Download full log” button instead of being truncated.Bug Fix. Start hooks that produce more than 1 MB of output no longer fail with a buffer error.
Automation Templates. The automations page now ships with starter templates (Code Review, DRY Code Check, No useEffects Check, Documentation Sync) tagged by category. Click a template to preview it, then Use this template to open the create form pre-populated. Read the docs!Unified Automation View/Edit. Viewing an automation now uses the same layout as editing, so the title, triggers, repository, instructions, and lifecycle are easier to scan. Action buttons (enable, run, edit, delete) live in the header.Grouped GitHub PR Events. GitHub
pull_request.opened and pull_request.synchronize automation triggers now support a Group PR events option. When enabled, later opened or synchronize events for a PR that already has a workspace from a prior run of the same automation route to that workspace as a follow-up message instead of spawning a new one, mirroring the Slack thread grouping behavior. Read moreUI Polish. Diff operation rows in the agent chat now match the layout of other event blocks (Command, Tool, Files, etc.): file icon first, title-case label, path, and a right-aligned chevron for expandable diffs.Copy Plan Markdown. The dashboard plan viewer now has a copy button next to the plan header that copies the raw markdown of the selected plan to your clipboard.Bug Fix. When multiple Slack automations fire on the same root message, thread replies now fan out to every triggered workspace instead of only one.Bug Fix. You can no longer add two automation triggers for the same event (e.g. two GitHub
pull_request.opened triggers), which previously caused the automation to fire twice.PR Closed Automation Trigger. Automations can now trigger when a pull request is closed, in addition to opened and updated.Opus 4.7 Support. Updated the Claude Agent SDK from 0.2.85 to 0.2.112, which includes support for the Opus 4.7 model. Workspaces using Claude will now use Opus 4.7 when the
opus model is selected.Copy and Move Infisical Between Scopes. Admins can now copy or move an Infisical connection from one environment scope to another (Global, Repository, or Repository Set) without re-entering credentials. Credentials are re-encrypted server-side so the Client Secret never leaves the vault. Read the docs!Bug Fix. The workspace state viewer now filters Infisical secrets and environment variables to the workspace’s actual scopes, matching what the engine receives at workspace creation.Bug Fix. Slack-triggered agents no longer stop after sending the acknowledgment message — they now continue with the implementation.
Smarter Slack Thread Context. Follow-up messages in long Slack threads now send only new messages to the agent instead of the full thread history, improving context quality on long conversations.
Infisical Integration. Connect Infisical to sync environment variables from your Infisical project into workspaces. Configure Machine Identity credentials per environment scope (Global, Repository, or Repository Set) from the Integrations section in the Variables tab. Secrets are automatically fetched at workspace creation time. Manual environment variables take priority over Infisical secrets. Read the docs!File Drop Indicator. The workspace chat input now shows a visual overlay with a dashed border and upload icon when dragging files over it, making it clearer where to drop files.Improved PR Footers. Pull requests created by Replicas agents now include a formatted HTML footer with contextual links back to the originating Linear issue or Slack thread, along with creator name and email attribution. This replaces the previous plain-text footer approach with a consistent, nicely formatted block across all PR sources.
Configurable Thinking Level. New thinking level selector in the chat toolbar lets you control how much reasoning the agent applies. Four levels are available: Low, Medium, High, and Max — where Max maps to
xhigh for Codex. Also available via the Replica API and MCP thinking_level parameter. When omitted, Claude defaults to high and Codex defaults to medium.Docker Support in Workspaces. Agents can now start the Docker daemon and run docker and docker compose commands inside Replicas workspaces. Docker is pre-installed but the daemon does not auto-start; agents are guided to start it on demand via the replicas-agent skill.Cleaner Workspace Names. Workspace names from Slack and Linear integrations no longer include a source prefix. Source is now indicated by badges in the UI.Subagent Thinking Indicator. The subagent viewer now shows a “thinking…” indicator with a spinning loader at the bottom of the chat history while a subagent is still in progress, matching the indicator in the main chat view.Bug Fix. Relay no longer offers Codex as a subagent provider when Codex credentials aren’t connected.Bug Fix. Relay no longer parallelizes dependent subagents (e.g. backend and frontend), so each one sees the other’s output.Bug Fix. Fixed a crash on the automations page when trigger configuration data was missing.Subagent Viewer. The subagent viewer now streams in-progress conversations in real time, uses a vertical timeline layout for better readability, and shows subagents from all workspace chats at once.
Slack Automation Triggers. Automations can now trigger on new messages in Slack channels. Filter to specific channels, and optionally attach the posting user to the workspace for security policy enforcement. When triggered, Replicas reacts with an eyes emoji and posts a threaded reply with a workspace link. Read moreAttach to User for Automation Triggers. Both GitHub and Slack automation triggers now support an optional “Attach to user” setting. When enabled, Replicas resolves the triggering user to a Replicas user in your organization and attributes the workspace to them, enabling security policy enforcement. Read moreAPI Plan Gating. API access is now restricted to Developer plan and above. Hobby plan users see an upgrade prompt in Settings > API Keys and receive a 403 error if they attempt to create or use API keys.Team Workspaces View. The sidebar now includes a “Team Workspaces” view that shows dashboard workspaces created by other org members, with a “team” badge and creator tooltip on hover.Bug Fixes. GitHub
@tryreplicas comments on another org member’s workspace now resolve the correct coding agent, and PR descriptions use the correct path-based workspace link format.Clickable Metadata Pills. Metadata pills on workspace chat messages (PR numbers, issue IDs, usernames, files, etc.) are now clickable links that open the underlying resource on GitHub, Linear, or Slack. Source badges (e.g. “GitHub”, “Linear”, “Slack”) are also clickable. Old messages stored before this update will continue to render without links.PR Attribution. Pull requests created by Replicas agents can now be attributed to you instead of the Replicas bot. When your GitHub account is connected via OAuth, the
gh CLI inside workspaces uses your personal token so PRs show you as the author. This is off by default and can be toggled at the user and organization level in Settings. Admins can also enforce it org-wide via the “Require PR user attribution” security policy. Read moreStructured Automation Messages. Automation-triggered workspace messages now display structured metadata pills (automation name, event type, repository, PR number, and sender) with clickable links instead of rendering as raw JSON. Cron-triggered automations show only the relevant fields. Existing messages are unaffected.Workspace Sharing. Workspace owners can now share dashboard-created workspaces with other org members. Shared workspaces appear in a dedicated “Shared” view in the sidebar.Sidebar View Modes. The sidebar filter has been redesigned with five view modes: Owned, Shared, Automations, Integrations, and API. Multiple views can be selected simultaneously, with source badges on each workspace card.Replica API overhaul. The Replica API has been redesigned with a cleaner interface, new endpoints, and richer workspace state.Breaking changes:
POST /v1/replicanow usesenvironment_idinstead of repository selectors.- Response fields normalized to snake_case (
total_pages,has_more).
GET /v1/replica/repositoriesandGET /v1/replica/repository-setsto discover available targets.- Chat management: list, create, and delete chats within a workspace.
POST /v1/replica/:id/messageswith support forchat_id,model,images, andplan_mode.GET /v1/replica/:id/hooksto inspect warm and start hook logs.GET /v1/replica/:id/eventsfor real-time SSE streaming.
GET /v1/replica/:idnow returns chats, detailed repository statuses, and optional environment/integration info via?include=environment,diffs.GET /v1/replicasupports filtering byuser_id,source, andstatus.- API key auth no longer requires an organization header.
replicas-mcp to v0.2.0. Read the MCP docs!Billing Page API Usage Graph. The billing page now shows a daily API usage chart (in minutes) to the right of the Seats and API Usage summary cards. The graph only appears when there is API usage in the current billing period.CLI Automation Management. The Replicas CLI now includes a full replicas automation (alias auto) command group for managing automations from the terminal. You can list, inspect, create, edit, run, and delete automations without leaving the command line. All commands support both flag-based (scriptable) and interactive input modes. Read the CLI docs!CI Failure Response for External Checks. Replicas now automatically responds to CI/CD failures from external check providers (e.g. Mintlify, CircleCI, Buildkite) in addition to GitHub Actions workflows. Two new toggles in Organization Settings → CI Failure Response let admins independently control auto-response for GitHub Actions workflows and external CI checks. Read the GitHub docs!Scoped Environment Skills. Skills can now be enabled per repository or per repository set, in addition to globally. The Environment Hub shows skills count badges for each repo and repo set, and the drill-in views have a new Skills tab with search, enable/disable, and an “inherited from Global” section showing read-only global skills.Workspace Lifecycle Policies. Automations now support configurable lifecycle policies that control how long workspaces stay active and when they auto-sleep or delete.Per-Minute Metered Billing. Workspace usage is now billed per-minute instead of per-hour, giving teams more granular cost control.Other improvements:
- Agent connection status is now shown in the CLI.
- Workspace creator name is automatically added to PR descriptions.
- Workspace sidebar now shows up to 100 workspaces (previously 10).
- Fixed sidebar responsiveness across CLI and web.
Updated Terms of Service and Privacy Policy to reflect the per-seat + metered API/Automations billing model that has been live since the March pricing overhaul, and to formalize the two sanctioned programmatic channels: the public REST API and Automations.Automations docs. Added a dedicated Automations page covering triggers, plan limits, the circuit breaker, and metered billing.Billing docs refresh. Billing now includes concrete examples and a clearer explanation of how seats versus metered usage are counted.CLI clarification. CLI docs now state explicitly that the CLI is for interactive use by a human operator and counts toward your seat.
Introducing Automations. Define rules that automatically trigger Replicas workspaces based on events. Create automations from the dashboard to respond to GitHub, Linear, or Slack events with custom workspace configurations.
- Automations are gated to paid plans (Hobby plan users see an upgrade prompt).
- Trigger pills show tooltips describing each automation’s configuration.
- Added diff viewer to
replicas interactfor reviewing code changes inline. - Added Create PR button to the CLI Info panel.
Interactive CLI. A new terminal UI (
replicas interact) built with OpenTUI for managing workspaces from the command line. Features include:- Sidebar with j/k keyboard navigation (wrapping)
- Wake sleeping workspaces directly from the CLI
- Auto-focus input box after creating a new workspace
- Structured rendering for special events (Slack, Linear, GitHub messages)
@replicas and @replicas-connector are now supported as aliases alongside @tryreplicas for triggering workspaces from GitHub.Other improvements:- Plan files are now saved to a standard location for easier access.
- Fixed workspace ownership being lost on pull requests created by agents.
Slack Thinking Status. Replicas now shows a thinking indicator in Slack threads while the agent is working, powered by the Slack Agents API.Linear & Slack API in Workspaces. All workspaces now have access to both the Linear and Slack APIs, allowing agents to fetch issue context, update statuses, and interact with Slack threads.
- Warm hooks/pools UI is now gated behind Team+ plans.
- Fixed unnecessary loading states in certain workspace views.
- Fixed an issue where warm workspaces could lose their PR URL.
- Added error boundaries to prevent workspace crashes from malformed data.
- Workspaces triggered from Linear now use Linear’s suggested branch name instead of generating one.
- Auto-reply accounts can now be toggled off without fully removing them.
- Added error page styling improvements.
Inline Diffs — Claude Write and Edit tool calls now render inline diffs directly in the workspace chat, making it easier to review agent code changes at a glance.Sandbox Resource Tiers — Workspaces now support configurable sandbox sizes:
- Memory tiers: 8GB and 16GB
- Disk tiers: 10GB, 20GB, and 32GB
- Added option for admins to configure a default workspace target for Slack and Linear triggers.
- Repository sets can now be created during onboarding.
Hook Logs — Warm and start hook output logs are now stored when run and viewable from the dashboard, making it easier to debug provisioning issues.Environment Variable Management — Copy or move environment variables and files between scopes (global, repository, repository set).
- Bulk delete workspaces from the sidebar.
- Copy-to-clipboard icons on chat messages.
- Fixed workspace rename error not showing a toast notification.
- Refreshed the Environments UI with a cleaner layout for managing variables and files.
- Preview URLs can now be authenticated, with all URLs set with a random hash by default for security.
- CLI and API key usage are now tracked separately in billing.
Subagent Viewer — Monitor subagent activity in a dedicated viewer with a back button for navigation.Improved Workspace Sidebars — Refreshed sidebar layout for better workspace navigation.Pricing Overhaul — Updated pricing model for seats and API usage.
- Linear parent issues now automatically update to “In Progress” when a Replicas task starts.
- Fixed Claude OAuth token refresh to prevent cross-workspace invalidation.
- Fixed orphaned mock workspaces lingering in the sidebar.
- Admin role is now required to delete repositories via the API.
- Fixed GitHub repo listing pagination for organizations with many repositories.
- Smooth hover transitions on sidebar items.
Introducing Preview URLs — agents can now expose local ports as public URLs so humans can view running services directly in their browser.
- Ask Replicas to use the replicas skill to create previews.
- Run
replicas preview create <port>to get a public URL for any service running in a workspace. - Preview ports are visible in the workspace Environment popover with clickable links.
- Various mobile UI fixes.
- Updated the UI to feel denser and easier to follow.
- Fixed an issue where agent auth refreshes would not invalidate warm pools.
- Added a delete button for workspace items in the sidebar.
- Fixed an issue where warmed workspaces would always have
warm-poolchecked out.
Introducing Replicas V1!Today, we publicly announced Replicas V1!What we announced (which people reading this changelog would already know):
- Replicas is a background coding agent for engineering teams: assign a ticket, walk away, get a PR.
- Works from Linear, Slack, GitHub, or directly via API, with each workspace running in its own sandboxed VM against your real codebase.
- Workspaces support warm hooks, environment variables and files, and custom skills, with preview URLs and workflow automations on the way.
- Added Skills to Replicas environments, with support for searching from skills.sh and making them available across workspaces.
- Added stale warm pool handling so environment changes do not fully deplete pools, improving workspace spin-up latency.
Introducing Warm Hooks — pre-warm workspaces by running setup scripts ahead of time so new workspaces provision instantly.
- Configure warm hooks at the organization level (Team/Enterprise) or per repository via
warmHookinreplicas.json - Maintain a pool of pre-warmed workspaces per repository or repository set
- Warm pools auto-invalidate when environment variables or files change
- Default 5-minute timeout per command (max 15 minutes), pre-warmed workspaces expire after 24 hours
- Revamped the dashboard with a settings sidebar so configuration is easier to access as more options roll out.
- Fixed an issue where mentioning
@tryreplicaswould sometimes not trigger a response. - Added support for setting default branches per repository.
- Added configuration to auto-respond to PR comments from specific users or bots, useful for code reviewers like Greptile.
- Reduced errors when Replicas automatically selects a repository or repository set.
- Updated chat rendering for incoming Slack, Linear, and GitHub messages to improve clarity.
- Added OpenAI API key support.
- Workspaces now support multiple repositories in a single environment.
- Added Repository Sets so teams can save and reuse multi-repo workspace presets.
- Added multiple chats per workspace for parallel or separated task threads.
- Added dashboard file uploads in chat so you can attach files directly to agent conversations.
- Added Environment settings for both variables and files, with scopes for global, per-repository, and per-repository-set configuration.
- Added an improved Diff Viewer in the workspace UI for reviewing code changes.
- Replicas workspaces now support Docker containers for development workflows.
- Added Claude Opus 4.6 model.
- Added GPT-5.3 Codex model.
- Slack threads linked to GitHub pull requests now mark messages with a checkmark when the PR is merged.
- Fixed Slack messages not waking sleeping workspaces.
- Repositories are now associated with only one organization. This also means that a
replicas.jsonfile withorganizationIdis no longer needed. - Added user and organization preferences to control whether workspaces close when linked pull requests close or merge.
- Fixed wake buttons staying in a pending state when switching between sleeping workspaces.
- Turns for both Codex and Claude Code can be interrupted. Messages that are queued are relayed back.
- Removed GPT Codex Mini from the list of available Codex models as it is not available with ChatGPT/Codex subscriptions.
- Fixed issue where pull requests created by agents were not being tracked in the database.
- Slack mentions now map sender email to git co-author details when available.
Workspaces can now be renamed inline. Click the workspace name in the chat header to edit it.
Admins can now change member roles directly from the Members settings page. Promotes members to admin or demotes admins to member, with safeguards against demoting the last admin.
Misc fixes for dashboard QOL:
- Fixed agent chat to have proper scrolling.
- Updated awoken workspaces to properly fetch plan.
Slack integration now supports image attachments. When you mention
@Replicas with an image attached, the agent can now see and process the image.Fixed regression with Claude Code authentication for AWS Bedrock in workspaces.Misc:
- Diff and Plan mode viewers now take up the full main area by hiding the chat input
Added Slack integration for triggering Replicas from Slack channels.
Read the docs!
- Connect Slack workspace via OAuth in organization settings
- Mention
@Replicasin any channel to create a workspace - Interactive dropdown to select repository when multiple are available
- Use
[repo:name]syntax to specify repository inline (inclusive matching) - Agent replies directly in the thread when finished
- Full thread context is passed to the agent for ongoing conversations
- Fixed issue where “Claude is thinking…” would persist indefinitely when using Plan Mode even when done
- Disabled user-input tools (ExitPlanMode, AskUserQuestion) for background agents
Introducing Plan Mode!Plan Mode lets agents explore the codebase and propose a plan before making changes. Plans are shareable and visible in the dashboard, and you can trigger Plan Mode by adding
/plan in GitHub, Linear, or dashboard chat.
Read here!Additional changes:- Linear no longer has a ‘middleman’ agent, instead spawning the workspace immediately.
- Added organization setting to set the default coding agent between Claude and Codex. Try it here here.
- Codex system prompts added by Replicas are inserted to
config.toml, instead of prepending to the message. - Fixed bug where sleep/wake could create multiple active workspace sessions
- Fixed bug where Codex would sometimes fail to trigger from Linear.
Workspaces now start 90% faster than before, thanks to Daytona sandboxes!
As the first major update approaching Replicas V1, we migrated our sandbox infrastructure to Daytona (daytona.io) for sandboxes!
Through our benchmarks, our replica creation -> agent working time went from ~1min to 5 seconds!No changes are required on your end.
- Added GPT-5.2 Codex model support for the Codex agent
- Updated Developer plan pricing from 50/month
- Added support for triggering Replicas tasks from GitHub issue comments (not just PR comments)
- Workspaces can now be created directly from GitHub issues by mentioning
@tryreplicas - When triggered from an issue, Replicas will create a PR and reply to the issue with the PR link and summary
- Issue workspaces are automatically cleaned up when the issue is closed
- Added validation to prevent replica names with spaces in the CLI, MCP, and API
- Fixed CLI build to avoid duplicate shebangs that crashed
replicas loginin Node 20+.
- Added streamable HTTP MCP at
https://api.tryreplicas.com/v1/mcp - Added stdio MCP client package (
replicas-mcp). - Added CLI commands for replica management:
list,get,create,send,delete,read - Added CLI command for listing repositories:
repos list
- Added a public alpha for the Replicas API. See docs.
- Fixed bug where comments on new PRs would spawn a new replica instead of routing to the original workspace when a replica created multiple PRs on different branches
- Workspaces triggered from GitHub PRs will now automatically receive notifications when CI/CD checks fail. The agent will investigate the failure, fix the issues, and push updates.
- Workspaces triggered from GitHub PRs will now automatically close when the associated PR is closed or merged.
You can now trigger Replicas on any pull request by mentioning
@tryreplicas in a PR comment, even if the PR wasn’t created by Replicas. When triggered, Replicas will:- Verify the repository belongs to your organization
- Create a new workspace to work on the PR
- Checkout the PR branch and address your request
- Automatically associate the workspace with the PR for future interactions
- Hobby: 24 hours
- Developer: 72 hours
- Team: (custom)
- Start hooks now run asynchronously and don’t block the engine from receiving messages
- Start hook outputs are logged to
~/.replicas/startHooks.logfor visibility - Start hooks only run once per workspace lifecycle (persisted across sleep/wake cycles)
- Coding agents are informed when start hooks are running with instructions to check the log
Sending messages to sleeping workspaces will no longer require you to ‘come back’, instead just queueing the message.
Linear Integration Enhancements & Organization Features
- Invite members to organizations for team collaboration
- Coding agents now respect pull request templates in repositories
- Purchase credits directly from the dashboard in flat amounts
- Workspaces now sleep after 1 hour of inactivity and permanently delete after 24 hours
- Messages can queue in the engine while agents are working
replicas.jsonnow supportsstartHookfor running shell scripts on engine startup and customsystemPromptfor agent instructions- Update credentials to connect your own GitHub, Claude Code, and Codex accounts—enables workspaces to use personal credentials instead of organizational defaults
- Fixed bug where Claude and Codex authentication would expire on sleeping workspaces
- Coding agents now accept images in messages
- Dashboard chat supports image uploads
- Linear issues with images in descriptions are automatically passed to agents
- CLI now displays notifications when updates are available
- Fixed issue where CLI connection to workspaces would timeout silently without warnings
- Workspaces that wake from sleep now properly persist thread and branch state
- Organization names can now be edited in settings
The Linear Agent integration is now available! Connect it to your Linear workspace in settings to assign tasks directly to Replicas. When you assign a task, Replicas will automatically spin up a workspace and create a PR for you.
Added GPT-5.1 and GPT-5.1 Codex.
Claude 4.5 Opus and GPT-5.1 Codex Max are now live in Replicas. Switch to the refreshed model selector to try the newest Claude release and OpenAI’s flagship Codex deployment with improved reasoning depth out of the box.
Replicas v0 is officially live! We’re making it easier to spin up production-grade development environments in minutes.GitHub-connected templates
- Connect and authorize your GitHub account once, then build workspace templates straight from your repositories!
- Workspaces inherit your Git credentials automatically, so git operations just work on first boot!
- Chat with Codex directly from the dashboard without opening a terminal!
- Active indicators now surface which workspaces are online! The dashboard is still evolving, and you’ll see rapid improvements over the next few weeks.
- Capture any workspace as a reusable snapshot and boot future workspaces from that image!
- Ideal for large project installs (think
node_modulesor language toolchains) so you only pay the setup cost once.
- Run
replicas codex-authto sync your Codex credentials with Replicas once. All future workspaces inherit that authentication automatically, so you can start chatting immediately.
Updated instance types for better performance:
- Standard: 2vCPU, 4GB RAM @ 2 credits/hour
- Large: 2vCPU, 8GB RAM @ 4 credits/hour
- Power: 4vCPU, 16GB RAM @ 8 credits/hour
We’ve replaced GitHub authentication via personal access token with an official GitHub connection. This lets you automatically scope access to your own repositories.Unfortunately, old templates are no longer viable after this change, and should be replaced.
You can configure your GitHub connection under Connections.
RMI v0.1.2:
- Fixed an issue in the Replicas Machine Image where Docker would not start correctly and block SSH access.
- New
--skip-authflag - Added--skip-authoption toreplicas connectandreplicas codecommands to skip copying Codex authentication to workspaces when not needed
npm i -g replicas-cli@latestWe’re excited to announce that all workspaces are now built on the Replicas Machine Image (RMI) - a standardized base VM that comes pre-configured with essential development tools and dependencies. Now, tooling comes pre-installed, making it faster and easier to get started with your projects.Other Updates
- New
replicas toolscommand - View the complete list of tools and dependencies available in RMI - Automatic agent setup - Codex, Claude Code, and Replicas CLI are now installed automatically on workspace setup
- Python 3.12.3
- pip 24
- Poetry 2.2.1
- uv 0.9.5
- Node.js 22.21.0
- npm 10.9.4
- nvm 0.40.3
- Yarn 1.22.22
- pnpm 10.18.3
- ESLint 9.38.0
- Prettier 3.6.2
- gcc 13.3.0
- g++ 13.3.0
- make 4.3
- Perl 5.38.2
- git 2.43.0
- vim 9.1.697
- nano 7.2
- curl
- wget
- tmux
- Docker Engine
npm i -g replicas-cli@latestWe’ve greatly simplified and fixed a bug with Codex authentication where tokens would not refresh as templates became too old.Replicas CLI v0.2.7
- Automatic Codex authentication - The CLI now automatically copies
~/.codex/auth.jsonto your workspace when you connect, eliminating manual setup - Auto port forwarding - Port 1455 is now tunneled automatically, so you can login from the workspace.
- Simplified templates - Codex CLI is now installed on every workspace by default. Templates no longer require agent authentication configuration
npm i -g replicas-cli@latestOther changes:- The beta is no longer closed! We’ve added several more abuse protection layers, allowing us to be confident in a general availability much earlier than expected.
- Fixed build issue with the CLI where tslib was missing.
We’re excited to announce that Replicas has officially launched into closed beta! We’re onboarding new signups on a rolling basis.Replicas CLIWe’ve released the Replicas CLI, making it super frictionless to work with your workspaces:
- Automatic file synchronization - Copy your development files and environment variables seamlessly
- SSH connectivity - Connect directly to your workspaces via SSH
- IDE integration - Open your workspaces directly in your favorite IDE
- Port tunneling - Easy access to services running in your workspaces