Delivery plan
Work is grouped by horizon rather than by release date. Progress counts milestones that already exist in the source tree — not estimates — so a bar only moves when code lands. Scope and priorities can change as implementation and feedback progress.
See how features connect on the roadmap, or browse every release on GitHub for full notes and downloads. Last reviewed .
- Now
- 6
- Next
- 10
- Later
- 35
- Milestones done
- 33 / 181
- Shipped
- 57
Now In progress
Being built or improved right now.
Agent code sandboxes
Let evaluated agents run code in an isolated sandbox: the built-in script sandbox today, an opt-in container next.
3 of 4 milestonesWhat’s done and what’s left
Today: The QuickJS provider runs agent JavaScript on desktop and in the CLI, with no network, filesystem or timers.
- Done: Sandbox provider contract and registry
- Done: Script sandbox provider on desktop and CLI
- Done: Threat model and design record
- To do: Opt-in local container provider on desktop#1055
Built on Agent evaluation suites
Try it from the docs
Open each protocol example from the docs straight into Restura against the public echo server.
3 of 4 milestonesWhat’s done and what’s left
Today: Each protocol page offers a sample collection for the public echo server and an Open in Restura desktop link; the web app imports the downloaded file by hand.
- Done: Public echo server
- Done: Per-protocol sample collections
- Done: Open-in-Restura links on desktop
- To do: One-click open in the web app
Contract testing
Attach an OpenAPI spec to a collection to generate mock routes today; validating live responses against it is being wired in.
3 of 5 milestonesWhat’s done and what’s left
Today: Collections load and check an OpenAPI spec and turn it into mock-server routes. A tested response validator exists but nothing calls it yet, and the collection settings say so.
- Done: Attach and validate an OpenAPI spec per collection
- Done: Generate mock-server routes from the spec
- Done: Response validator library with tests
- To do: Validate responses on send and show violations in the response panel
- To do: Report contract results in collection runs and CLI reports
Built on HTTP / REST
Protocol & IPC coverage
Strengthen meaningful regression coverage at protocol and desktop IPC boundaries.
3 of 5 milestonesWhat’s done and what’s left
Today: The shared protocol core enforces coverage floors, a desktop end-to-end suite runs in CI, and a global uncovered-code budget can only shrink.
- Done: Coverage floors on the shared protocol core
- Done: Desktop end-to-end suite in CI
- Done: Global uncovered-code budget that only ratchets down
- To do: Per-directory floors for desktop IPC and Worker handlers
- To do: Contract tests for IPC channel and validator parity
Accessibility improvements
Improve keyboard and screen-reader workflows, particularly the workflow builder and response viewer.
2 of 5 milestonesWhat’s done and what’s left
Today: The October UX overhaul shipped a type-scale ratchet, labelled console controls, and announced responses. Automated checks and keyboard use of the workflow canvas are still missing.
- Done: Type scale and contrast pass (#789)
- Done: Labelled console buttons and announced responses (#764)
- To do: Automated axe checks for the workflow builder and response viewer
- To do: Keyboard-operable workflow canvas
- To do: Live regions for streaming responses
Secret handle management
Bring the remaining credential fields onto desktop secret handles and manage them in one place.
1 of 4 milestonesWhat’s done and what’s left
Today: The handle store and send-time resolution are in place, and fields migrate one at a time. Importers do not yet offer conversion, and there is no list of stored handles.
- Done: Handle store resolved only at send time
- To do: Offer to convert imported secrets into handles
- To do: Settings panel listing and deleting handles
- To do: Migrate the remaining credential fields
Built on Desktop secret handles
Next Planned
Scoped and intended, with the groundwork in place.
Postman round-trip follow-ups
Export to Postman from the CLI, keep saved responses as examples, and open Postman v3 folders as live collections.
2 of 5 milestonesWhat’s done and what’s left
Today: Postman v2.x / v3 import and v2.1 / v3 export work in the app. Saved responses live only in the _postman sidecar, and the CLI cannot export. WebSocket / Socket.IO items import and export as native collection items.
- Done: Postman v2.x / v3 import and v2.1 / v3 export in the app
- To do: CLI export to Postman (--format postman-v2.1|postman-v3)#1204
- To do: Saved responses / examples model (today the _postman sidecar only)#1205
- To do: Open Postman v3 folders as live git-native collections#1206
- Done: Native WebSocket / Socket.IO collection items (#1209)
Built on Postman v2 / v3 import and export
Standalone environment export
Export an environment independently of a collection.
2 of 5 milestonesWhat’s done and what’s left
Today: Environments travel inside OpenCollection collection exports, and Postman environment files can be imported. There is no way to export one environment on its own.
- Done: Environments included in collection exports
- Done: Postman environment import
- To do: Export a single environment as Restura or Postman JSON
- To do: Choose whether secret values are included
- To do: Round-trip test against the importer
AI assistant on web
Bring request-aware assistance to the browser with careful API-key handling.
2 of 5 milestonesWhat’s done and what’s left
Today: Provider streaming, decoding, and redaction live in the backend-agnostic protocol core, and the chat panel exists. Only the desktop app has a transport for it.
- Done: Provider decoders and redaction in the shared core
- Done: Request-aware chat panel
- To do: Decide how browser users supply and protect API keys
- To do: Streaming /api/ai route with SSRF and rate limits
- To do: Web transport alongside the desktop bridge
Built on Request-aware assistant
WebSocket headers on web
Send custom handshake headers from the browser by relaying connections through the Restura server.
1 of 3 milestonesWhat’s done and what’s left
Today: The server-side ticket and relay routes exist behind the SSRF gate, but the web app still connects directly, so custom headers are dropped with a warning.
- Done: Ticketed relay routes on the Worker and self-hosted server
- To do: Web app connects through the relay when headers are set#898
- To do: Header policy and end-to-end tests
Built on WebSocket
Passphrase-protected web storage
Encrypt the web app’s local data with a passphrase only you know.
1 of 4 milestonesWhat’s done and what’s left
Today: A passphrase key provider is implemented and documented as planned; there is no Settings screen to set or unlock it yet.
- Done: Passphrase key provider in the storage layer
- To do: Settings → Security passphrase setup and unlock prompt
- To do: Re-encrypt existing browser data, with a recovery warning
- To do: Security regression tests
Built on Local-first storage
PAC proxy scripts
Route desktop traffic using your organisation’s proxy auto-config script.
1 of 4 milestonesWhat’s done and what’s left
Today: Desktop networking has handler scaffolding, but proxy settings cannot choose PAC and the script is never loaded, so it is marked unsupported.
- Done: Handler scaffolding in desktop networking
- To do: PAC option in proxy settings
- To do: Load the script through the desktop session proxy
- To do: PAC evaluation tests
Built on mTLS & native networking
CLI protocol parity
Run more of what the app can send from CI: WebSocket requests, per-domain client certificates, and protobuf bodies.
1 of 4 milestonesWhat’s done and what’s left
Today: A standalone WebSocket executor already exists in the CLI but nothing routes to it; per-domain certificates and protobuf bodies are documented as not yet supported.
- Done: Standalone WebSocket executor
- To do: Run WebSocket requests in collection runs#881
- To do: Honour per-domain client certificates
- To do: Protobuf request bodies
Built on CLI collection runner
WebSocket message scripts
Extend scripting to incoming WebSocket messages.
1 of 5 milestonesWhat’s done and what’s left
Today: The bounded QuickJS sandbox that runs HTTP scripts is shared and ready; the WebSocket client has no script hook yet.
- Done: Bounded script sandbox shared across protocols
- To do: On-message script saved with WebSocket requests
- To do: Run per incoming frame with time and rate limits#899
- To do: rs.message and rs.ws.send scripting APIs
- To do: Script output and errors in the message log
Built on WebSocket, Sandboxed scripts
MCP server: full coverage
Extend the agent tools to the rest of Restura: connection presets, workflows, exports and more read tools.
Not startedWhat’s done and what’s left
Today: Agents can build, edit, delete and send HTTP, GraphQL, gRPC, SSE and MCP requests, environments and imports. The items below are not exposed to agents yet (ADR 0041, Scope).
- To do: WebSocket, Socket.IO, Kafka and MQTT connection presets, with keychain-safe credentials
- To do: HAR import, export_collection and deleting environments
- To do: Workflows, active tab, last response and capabilities tools
- To do: Inline (unsaved) sends and MCP resources and prompts
- To do: Revert a whole agent session at once, and an "agent connected" status indicator
Built on Restura as an MCP server
Decision models
Test decision models such as Cloudflare Clef and TypeSafe Jev, which answer typed questions with calibrated probabilities, and use them as fast eval scorers.
Not started
Later Exploring
Ideas being weighed. These are not commitments.
Code snippets for every protocol
Generate code for WebSocket, SSE, MCP, GraphQL, and gRPC requests the way HTTP does today.
1 of 3 milestonesFirst steps
Today: Generators for WebSocket, SSE, MCP, and GraphQL are written but nothing in the interface uses them.
- Done: WebSocket, SSE, MCP, and GraphQL generators
- To do: Snippet panel on each protocol#916
- To do: grpcurl and buf curl output
MQTT request and response
Publish with a reply topic and correlation ID, wait for the matching reply, and show round-trip time.
1 of 3 milestonesFirst steps
Today: MQTT 5 reply topic and correlation data can be set on publish; nothing matches replies.
- Done: Reply topic and correlation data on publish
- To do: Automatic reply subscription
- To do: Reply matching and latency
Broker connection presets
Save Kafka and MQTT connections, topics, and payload templates in collections to share them through Git.
1 of 3 milestonesFirst steps
Today: Kafka and MQTT connections are saved locally with secrets in secure storage, but cannot go in a collection.
- Done: Secure storage for broker secrets
- To do: Connection item in collections#886
- To do: Collection file format support
Unified streaming timeline
See WebSocket, Socket.IO, SSE, Kafka, and MQTT traffic in one filterable, exportable timeline.
1 of 4 milestonesFirst steps
Today: The console already mirrors WebSocket and SSE frames; each protocol otherwise keeps its own message log.
- Done: Console mirrors WebSocket and SSE frames
- To do: Shared timeline event model
- To do: Kafka, MQTT, and Socket.IO sources
- To do: Filter, correlate, and export
Stream assertions
Assert on streamed frames and events in the app, such as a match within the first N events.
1 of 4 milestonesFirst steps
Today: The CLI exposes SSE stream events to test scripts; in the app, scripts run for HTTP, GraphQL and every gRPC call type (gRPC stream tests see all messages via rs.response.messages).
- Done: CLI stream-event script context
- To do: In-app stream script hook#899
- To do: Assert-within-N-events helpers
- To do: WebSocket, MQTT, and Kafka sources
Built on WebSocket
MCP server tests in CI
Save MCP calls in collections with test scripts and run them from the CLI.
1 of 4 milestonesFirst steps
Today: The CLI can send MCP requests, but saved collections keep only a ping; the app runner skips MCP.
- Done: Pooled JSON-RPC runner
- To do: Method and arguments saved in collections#877
- To do: In-app runner support#899
- To do: Test scripts on MCP results
Built on MCP client, CLI collection runner
MCP conformance report
Run a server against older and newer protocol versions and report differences and schema violations.
1 of 5 milestonesFirst steps
Today: The desktop client negotiates protocol versions and the SDK validates structured tool output.
Self-hosted audit logging
Explore an operational audit trail for self-hosted deployments.
Not startedFirst steps
Today: Nothing yet. The self-hosted Node server and its middleware are the natural place to start.
- To do: Decide which events matter and how long to keep them
- To do: Structured audit events from the self-hosted server
- To do: Export to an operator-owned log sink
Scheduled test runs
Explore running tests on a schedule rather than starting each run manually.
Not startedFirst steps
Today: The CLI already produces JUnit, HTML, and JSON reports, so CI schedulers can run collections today.
- To do: Documented GitHub Actions and cron recipes
- To do: Decide whether an in-app scheduler is worth it
- To do: Run history and failure notifications
Built on CLI collection runner
Natural-language requests
Explore turning a plain-language description into a request.
Not startedFirst steps
Today: The desktop assistant and AI Lab already generate requests from OpenAPI, which this would build on.
- To do: Prototype prompt-to-request with the desktop assistant
- To do: Review-before-send flow that never executes silently
- To do: Evaluate quality with AI Lab datasets
Risk-scored tool approvals
Explore letting a decision model score the risk of an agent tool call before asking you to approve it.
Not startedFirst steps
Today: Every non-read agent tool call needs a manual approval.
- To do: Design record and security review#1056
- To do: Risk questions over the pending call#1056
- To do: Auto-approve or escalate by threshold#1056
Built on Agent evaluation suites
Model drift watch
Explore re-running saved evals on a schedule and alerting when a provider update lowers your pass rate.
Not startedFirst steps
Today: Eval reports compare a run against the previous run of the same eval, on demand.
- To do: Per-eval pass threshold#1024
- To do: Scheduled eval runs on desktop#1053
- To do: Regression alerts#1053
Built on AI Lab, Scheduled test runs
API explorer agent
Explore an agent that probes an API from its schema and saves each failure as a request and an eval case.
Not startedFirst steps
Today: Agents can call saved requests as tools, and evals can be generated from OpenAPI specs.
- To do: Schema-driven probing with approvals#1054
- To do: Failures saved as requests#1054
- To do: Failures saved as eval cases#1054
Built on Agent evaluation suites
JetBrains integration
Explore in-editor request execution for JetBrains IDEs.
Not startedFirst steps
Today: No JetBrains code exists. The VS Code extension and CLI show the shape it could take.
- To do: Gauge demand from users
- To do: Reuse the CLI as the execution engine
- To do: Minimal plugin that runs OpenCollection files
Self-hosted scale-out
Explore running Restura’s server across several instances: shared rate limits, a Helm chart, and upstream mTLS.
Not startedFirst steps
Today: Single-instance self-hosting ships today; these are listed as out of scope for v1 and open to contributions.
- To do: Reference Helm chart
- To do: Pluggable shared rate-limit store
- To do: Optional upstream mTLS
Built on Self-hosted Docker
NATS & JetStream
Publish, subscribe, and make request-reply calls against NATS, including JetStream consumers.
Not startedFirst steps
Today: Kafka and MQTT clients ship on desktop behind broker address guards; NATS has no client yet.
- To do: Core publish and subscribe
- To do: JetStream consumers
- To do: TLS and credentials authentication
AMQP / RabbitMQ
Browse exchanges and queues, and publish or consume AMQP 0-9-1 messages.
Not startedFirst steps
Today: No AMQP client exists yet; the broker connection guards used by Kafka and MQTT would apply.
- To do: Connect and publish
- To do: Consume with acknowledgements
- To do: Exchange and queue browser
HTTP/3 & WebTransport
Send requests over QUIC and open WebTransport sessions, showing the negotiated protocol.
Not startedFirst steps
Today: Desktop HTTP records the negotiated ALPN protocol; there is no QUIC transport.
- To do: HTTP/3 requests
- To do: Alt-Svc negotiation
- To do: WebTransport streams
1Password & Vault
Add 1Password and HashiCorp Vault alongside the cloud secret providers.
Not startedFirst steps
Today: AWS, Google, and Azure secret providers ship; the provider list is a closed set in the profile schema.
- To do: Pluggable provider interface
- To do: 1Password adapter
- To do: HashiCorp Vault adapter
Workspace trust
Ask before running scripts or agent tools from a freshly cloned collection.
Not startedFirst steps
Today: Scripts run in a bounded QuickJS sandbox; there is no per-workspace trust decision.
- To do: Trust model
- To do: Desktop trust prompt
- To do: CLI trust flag
AsyncAPI import
Turn AsyncAPI channels into Kafka, MQTT, WebSocket, and SSE requests.
Not startedFirst steps
Today: Collections can record an AsyncAPI spec kind, but no importer reads AsyncAPI documents.
- To do: Spec parser
- To do: Channel-to-request mapping
- To do: Server and binding variables
Account-free team sync
Share collections and environments with teammates through an encrypted relay you host, with no cloud account.
Not startedFirst steps
Today: Teams share collections through Git today; there is no live sync.
- To do: Sync design record
- To do: Encrypted relay
- To do: Conflict resolution view
Visual assertion builder
Add status, JSONPath, schema, and latency checks without writing a script.
Not startedFirst steps
Today: Assertions are written as scripts in the sandbox; there is no no-code builder.
- To do: Assertion schema
- To do: Builder interface
- To do: CLI report parity
Mocks from recordings
Generate mock routes from recorded traffic and request history, with replay matching rules.
Not startedFirst steps
Today: The mock server serves hand-written and OpenAPI-generated routes only.
- To do: Recordings to routes
- To do: Matching rules
- To do: Record and replay mode
Uptime monitors
Run collections on a schedule from a self-hosted server and alert on failures.
Not startedFirst steps
Today: Collections run on demand in the app and the CLI; nothing runs on a schedule.
- To do: Monitor definitions
- To do: Alert channels
- To do: Run history dashboard
AI tests from traffic
Draft assertions and regression suites from recorded traffic, with secrets redacted first.
Not startedFirst steps
Today: AI Lab generates eval cases from OpenAPI specs and history; request test suites are written by hand.
Plugin API
Extend Restura with sandboxed plugins for custom auth, importers, and panels.
Not startedFirst steps
Today: Extension points are internal; scripts already run in a QuickJS sandbox a plugin host could reuse.
- To do: Manifest and permissions
- To do: Sandboxed host interface
- To do: Local plugin loader
Saved message templates
Save reusable WebSocket, Socket.IO, and MQTT messages with variables and send them in one click.
Not startedGraphQL operation picker
Pick an operation from a multi-operation document and send it over the right transport automatically.
Not startedFirst steps
Today: The operation type and name are read with pattern matching, and the first operation is always sent.
- To do: Parse documents into operations#870
- To do: Operation chips in the editor
- To do: Route subscriptions to WebSocket automatically
gRPC schema drift check
Save the reflected schema with a request and flag breaking changes the next time you reflect.
Not startedFirst steps
Today: Reflection fetches full descriptor sets, but nothing stores them on the request or compares versions.
- To do: Schema snapshot saved on requests#871
- To do: Descriptor diff
- To do: Breaking-change warnings
Built on Contract testing
Unknown-field warnings for gRPC
Warn before sending when request fields do not exist in the message type, instead of dropping them.
Not startedFirst steps
Today: Unknown request fields are silently dropped when the message is encoded.
- To do: Detect dropped fields
- To do: Pre-send warning
- To do: Strict mode setting
Read-only Kafka topic browser
Peek at a topic from a time or offset without creating a consumer group or committing offsets.
Not startedFirst steps
Today: Every consume joins a consumer group, so browsing a topic creates the group and commits offsets.
- To do: Group-free partition assignment#930
- To do: Start from time or offset
- To do: No-commit browsing mode
Schema Registry subject picker
Choose a subject and version instead of typing a schema ID, and see the schema beside decoded records.
Not startedFirst steps
Today: Producing with Schema Registry needs a numeric schema ID.
- To do: Subject and version browser#926
- To do: Latest-version produce
- To do: Schema shown beside records
Import MCP client configs
Add servers from the mcp.json files used by Claude Desktop, Cursor, and VS Code.
Not startedFirst steps
Today: MCP servers are added one by one by URL.
- To do: mcp.json parser
- To do: Remote server import
- To do: Local command servers after stdio support#854
Nothing matches these filters.
Recently shipped
- Workflow canvas undo/redo, full screen, and theming#792 · Automation & Testing
- Opt-in sample collection and drop-anywhere import#793 · Collections
- Globals editor, scope inspector, and folder variables#791 · Collections
- Response viewer: timing breakdown, JSON tree, compare, and cancel#785 · Protocols
- “Open as” protocol switch and resizable SSE/WebSocket settings#788 · Protocols
- Type-scale ratchet and accessibility fixes#789 · Platforms & Integrations
- Daily-driver URL bar, variables, navigation, and editors#780 · Platforms & Integrations
- Variables resolved inside request bodies#779 · Protocols
- AI prompt caching, cache-aware cost, and latest models#766 · AI & Agents
- GraphQL introspection and OAuth 2.0 refresh through the shared executor#772 · Protocols
Help shape the plan
Section titled “Help shape the plan”Suggest a feature with your use case, the platform you use, and the outcome you need. Items under Later move forward fastest when someone describes a concrete need.