fix(model): keep provider/model in sync when switching models by chindris-mihai-alexandru · Pull Request #2873 · antinomyhq/forgecode

chindris-mihai-alexandru · 2026-04-06T22:08:44Z

Summary

This PR fixes a model switching bug where /model could select a model from a different provider but keep the previous provider active, leading to invalid provider/model pairs and runtime 404s.

Changes

crates/forge_main/src/ui.rs
- select_model now returns (ProviderId, ModelId) instead of only ModelId
- /model now updates provider+model atomically when the selected model belongs to a different provider
- preserves same-provider behavior (model-only update)
- improves cursor restore by matching provider+model pair
crates/forge_services/src/app_config.rs
- set_default_provider_and_model now updates the in-memory session cache after persisting
- adds test: test_set_default_provider_and_model_updates_session_cache
crates/forge_services/src/agent_registry.rs
- agent registry config is now refreshable instead of a fixed startup snapshot
- reload_agents() refreshes config from disk/env before rebuilding agents
- adds test: test_reload_agents_refreshes_provider_model_from_config_file

Repro (before)

Set default provider to Modal
Use /model to select a model from another provider (e.g. Claude)
UI reports model switched, but requests are sent with stale provider/model pairing
Modal returns 404 Unknown model

Behavior (after)

Selecting a cross-provider model via /model now switches both provider and model together
Active agent/default config stay in sync with persisted config

Validation

cargo check -p forge_main -p forge_services
cargo test -p forge_main -- --nocapture
cargo test -p forge_services -- --nocapture
cargo test -p forge_api -- --nocapture

Co-Authored-By: ForgeCode noreply@forgecode.dev

Add AdaL CLI by SylphAI as a new provider option in ForgeCode. AdaL is registered as an OpenAI-compatible provider with api.sylph.ai endpoints, enabling ForgeCode users to access AdaL's models via API key auth. Changes: - Add 'adal' entry to provider.json with OpenAI response type - Add ProviderId::ADAL constant to forge_domain - Wire up display_name ("AdaL"), FromStr, and built_in_providers() - Add missing FromStr arms for vertex_ai_anthropic, bedrock, opencode_zen - Add unit tests for ADAL display name, from_str, and built_in_providers Co-Authored-By: ForgeCode <noreply@forgecode.dev>

Add Modal as a new provider in ForgeCode, enabling access to Z.ai's GLM-5 745B parameter model hosted on Modal's infrastructure. Modal offers an OpenAI-compatible API endpoint with free GLM-5 access until April 30th, 2026. Changes: - Add 'modal' entry to provider.json with GLM-5-FP8 model - Add ProviderId::MODAL constant to forge_domain - Wire up display_name ("Modal"), FromStr, and built_in_providers() - Add unit tests for MODAL display name, from_str, and built_in_providers Co-Authored-By: ForgeCode <noreply@forgecode.dev>

Combines both provider branches to enable AdaL (SylphAI) and Modal (GLM-5) support. Co-Authored-By: ForgeCode <noreply@forgecode.dev>

Ensure /model selection updates provider+model atomically when the chosen model belongs to a different provider, and refresh agent defaults from latest config on reload. Co-Authored-By: ForgeCode <noreply@forgecode.dev>

Copilot

Pull request overview

Fixes a bug where selecting a model from a different provider could leave the previous provider active, producing invalid provider/model pairs and resulting request failures.

Changes:

Update model selection flow to return (ProviderId, ModelId) and switch provider+model together when crossing providers.
Keep in-memory session config cache in sync after persisting provider/model changes.
Make the agent registry refresh its config on reload so provider/model changes apply immediately.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
crates/forge_services/src/app_config.rs	Updates the in-memory session cache after persisting provider/model changes; adds test coverage.
crates/forge_services/src/agent_registry.rs	Stores config in a runtime `RwLock` and refreshes config from disk/env during `reload_agents`; adds test.
crates/forge_repo/src/provider/provider.json	Adds new provider definitions for `adal` and `modal` (including a Modal-hosted model).
crates/forge_main/src/ui.rs	Changes model picker to select provider+model pairs and applies atomic updates when switching providers.
crates/forge_domain/src/provider.rs	Adds built-in provider IDs for `adal` and `modal` and updates parsing/display tests.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-06T22:12:40Z

crates/forge_services/src/agent_registry.rs

+    /// Reloads Forge config from defaults, global file, legacy file, and env.
+    fn load_current_config(&self) -> anyhow::Result<ForgeConfig> {
+        ConfigReader::default()
+            .read_defaults()
+            .read_global()
+            .read_legacy()


load_current_config layers config sources in a different precedence order than the rest of the codebase (e.g. ForgeConfig::read() uses legacy -> defaults -> global -> env). Calling read_legacy() after read_global() means legacy values can override the primary TOML config during reload_agents, leading to surprising provider/model resolution after switching. Align the read order with crates/forge_config/src/config.rs:279-285 (or call ForgeConfig::read()/a shared helper) so reload semantics match startup semantics.

Suggested change

/// Reloads Forge config from defaults, global file, legacy file, and env.

fn load_current_config(&self) -> anyhow::Result<ForgeConfig> {

ConfigReader::default()

.read_defaults()

.read_global()

.read_legacy()

/// Reloads Forge config using the same precedence as startup:

/// legacy file, defaults, global file, and env.

fn load_current_config(&self) -> anyhow::Result<ForgeConfig> {

ConfigReader::default()

.read_legacy()

.read_defaults()

.read_global()

Updated in 069e75a. load_current_config() now delegates to ForgeConfig::read() so reload order matches startup precedence exactly (legacy -> defaults -> global -> env).

Copilot · 2026-04-06T22:12:40Z

crates/forge_services/src/agent_registry.rs

+        let fixture_original_home = std::env::var("HOME").ok();
+        unsafe {
+            std::env::set_var("HOME", fixture_home.path());
+        }


This test mutates the process-wide HOME env var but only restores it at the end of the function. If an assertion/unwrap panics, HOME won’t be restored, which can cascade into unrelated test failures despite the mutex. Prefer an RAII guard (similar to EnvGuard in crates/forge_config/src/reader.rs) or scopeguard so env restoration happens in Drop even on early returns/panics.

Fixed in 069e75a. Added a HomeEnvGuard RAII helper in the test module that captures and restores HOME in Drop, and also holds the mutex lock for serialized env mutation. This guarantees cleanup even on panic.

…est guard Addresses Copilot review feedback by using ForgeConfig::read() for reload semantics and restoring HOME via RAII in tests. Co-Authored-By: ForgeCode <noreply@forgecode.dev>

chindris-mihai-alexandru and others added 4 commits April 6, 2026 21:34

Merge feat/adal-provider and feat/modal-provider into main

79ec314

Combines both provider branches to enable AdaL (SylphAI) and Modal (GLM-5) support. Co-Authored-By: ForgeCode <noreply@forgecode.dev>

Copilot AI review requested due to automatic review settings April 6, 2026 22:08

github-actions bot added the type: fix Iterations on existing features or infrastructure. label Apr 6, 2026

Copilot started reviewing on behalf of chindris-mihai-alexandru April 6, 2026 22:09 View session

Copilot AI reviewed Apr 6, 2026

View reviewed changes

fix(agent-registry): align config reload precedence and harden HOME t…

069e75a

…est guard Addresses Copilot review feedback by using ForgeConfig::read() for reload semantics and restoring HOME via RAII in tests. Co-Authored-By: ForgeCode <noreply@forgecode.dev>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(model): keep provider/model in sync when switching models#2873

fix(model): keep provider/model in sync when switching models#2873
chindris-mihai-alexandru wants to merge 5 commits intoantinomyhq:mainfrom
chindris-mihai-alexandru:fix/model-provider-switch-sync

chindris-mihai-alexandru commented Apr 6, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 6, 2026

Uh oh!

chindris-mihai-alexandru Apr 6, 2026 •

edited

Loading

Uh oh!

Copilot AI Apr 6, 2026

Uh oh!

chindris-mihai-alexandru Apr 6, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

chindris-mihai-alexandru commented Apr 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Repro (before)

Behavior (after)

Validation

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

chindris-mihai-alexandru Apr 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

chindris-mihai-alexandru Apr 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chindris-mihai-alexandru commented Apr 6, 2026 •

edited

Loading

chindris-mihai-alexandru Apr 6, 2026 •

edited

Loading

chindris-mihai-alexandru Apr 6, 2026 •

edited

Loading