bolt.gives

[embire2]

Hi All,

We have tried very hard to get the community admins who are supposed to be assisting us in adding new contributors to the app, but have been met with no response.

There is currently only 1 x Contributor that is working on the entire project and it is an uphill battle for him.
@DevFlex-AI and myself have started a new fork of the project where we will be rewriting basically the whole project as we want to make it as fast and easy to use as possible.

If you are interested, you can check out https://bolt.gives or visit the Git here: https://github.com/embire2/bolt.gives

We invite you to submit any bug issues you find and contribute where you can so that we can put bolt back on the map. We have already removed alot of bloat and added a bunch of features such as native web browsing. The app also allows you to provide it with API documentation, it will study it if instructed and will keep it in a .md file ready to analyze whenever you need it to build out an endpoint.

If admins of this repo come to the party, we do not mind moving the work we do on bolt.gives to this project as long as we can become contributors so that we can keep pushing forward.

In the meantime, we are on version 1.0.1 of bolt.gives and are working towards version 1.0.2, but need your valuable input.

@admins - We are even willing to completely drop the project if you join our efforts in making bolt shine.

Until next time,
E

[DevFlex-AI]

@embire2 @admins @embire I agree because only me and embire2 were working on this project so we could really use help

Until next time
D

[embire2]

Herewith the Roadmap to version 1.0.2 - https://github.com/embire2/bolt.gives:

Steps towards v1.0.2 - ETA: 7 days
Delivery Paths (P0)
Live Development Commentary Stream
Stream concise step-by-step commentary while coding (plan -> action -> verification -> next step).
Separate commentary from code/actions in protocol and UI.
Keep commentary visible while work is in progress (not only at completion).
Agent Anti-Stall + Auto-Recovery
Detect stuck states (repeated tool calls, no-progress loops, long inactivity/timeouts).
Add recovery strategies (retry/backoff, summarize-and-continue, explicit recovery prompt).
Surface recovery actions in the commentary timeline.
Execution Transparency Panel
Show selected model, tool invocations, active step, elapsed time, and token/cost estimate.
Show a short "why this action" reason for major agent steps.
Persist per-run execution trace for review.
Safer Autonomy Modes
Add explicit modes: read-only, review-required, auto-apply-safe, full-auto.
Add per-tool approval policy controls.
Display active autonomy mode in chat/workflow UI at all times.
Reliability Guardrails For Web/Tool Tasks
Keep tool schemas compatible with strict providers (including Codex-style strict function validation).
Add robust URL/tool input validation with clear fallback errors for users.
Add regression tests for schema compatibility across major model families.
Next Paths (P1)
Persistent Project Memory (Scoped)
Store lightweight persistent project summaries/architecture notes between sessions.
Use persisted memory to reduce repeated context rebuilding.
Sub-Agent Framework (Minimal)
Introduce a manager/worker pattern for long-running tasks.
Ship behind a feature flag.

[DevFlex-AI]

@embire2 have these features been added or should i add them or are we gonna add feature each of assigned a different feature

[embire2]

v1.0.2 has just dropped. This is available from: https://github.com/embire2/bolt.gives

v1.0.2 Planning (Playwright Research)

Date: 2026-02-16

Goal

Define a focused, high-impact feature set for v1.0.2 based on what developers are actively asking for in agentic coding products.

Delivery Status (Updated: 2026-02-17)

•  P0.1 Live Development Commentary Stream
  • Shipped in 970f1e3 (feat: stream live development commentary in execution timeline)
  • Commentary is now streamed as a separate protocol event and shown in timeline UI.
•  P0.2 Agent Anti-Stall + Auto-Recovery
  • Shipped in e81c073 (feat: add anti-stall recovery with auto-finalize fallback)
  • Detects repeated tool loops, no-progress streaks, and stream timeouts; applies backoff and recovery.
•  P0.3 Execution Transparency Panel
  • Shipped in c567b20 (feat: add execution transparency panel and autonomy enforcement)
  • Shows active model/provider, current step, elapsed runtime, token/cost estimate, recent tool calls, and "why this action".
•  P0.4 Safer Autonomy Modes
  • Shipped in c567b20 (feat: add execution transparency panel and autonomy enforcement)
  • Adds read-only, review-required, auto-apply-safe, full-auto with runtime enforcement and persisted mode selection.
•  P0.5 Reliability Guardrails for Web/Tool Tasks
  • Added cross-provider schema regression suite:
    • app/lib/.server/llm/tools/tool-schema-compatibility.spec.ts
    • tests/api.system.tool-schema-matrix.spec.ts
  • Added live compatibility matrix endpoint: /api/system/tool-schema-matrix
  • Standardized strict-compatible schema validation for web_search and web_browse across strict and standard provider profiles.
•  P1.6 Persistent Project Memory (Scoped)
  • Added scoped memory store and reuse pipeline:
    • app/lib/.server/llm/project-memory.ts
    • persisted client handoff via stream data event project-memory
  • Memory is injected back into build prompts for context reuse.
•  P1.7 Sub-Agent Framework (Minimal)
  • Added feature-flagged planner/worker flow in api.chat:
    • planner sub-agent pre-plans execution
    • worker agent executes with plan context
  • Controlled by BOLT_SUB_AGENTS_ENABLED=true.
• Validation gate for current main push:
  • pnpm run typecheck passed
  • pnpm run lint passed
  • pnpm test passed
  • pnpm run build:highmem passed
  • E2E smoke (strict OpenAI path on https://alpha1.bolt.gives) passed
  • E2E smoke (standard OpenAI model path on https://alpha1.bolt.gives) passed

How This Was Researched

Used Bolt.gives' Playwright-powered web browsing/search service to gather current requests from:

• Cursor Ideas forum
• Windsurf feature-request board
• OpenAI Codex GitHub issues
• Hacker News threads on coding agents
• GitHub Copilot public pages/discussions

What People Are Asking For (Patterns)

1. Better agent reliability and fewer stuck loops.

• Users report agents getting stuck, needing manual "continue", or looping.
• Evidence: Cursor Ideas/Feedback topics and Windsurf "Cascade error" / "continue" complaints.

2. Transparent execution: show what the agent is doing and why.

• Users want visibility into plan, progress, model/tool usage, and decision trail.
• Evidence: requests around planning mode, model selection visibility, and repeated community concerns about trust and boundaries.

3. Stronger repo-scale context and multi-repo workflows.

• Users want persistent memory/knowledge graph and cross-repo background work.
• Evidence: Cursor Ideas requests for persistent knowledge graph and multi-repository background mode.

4. Sub-agents/plugins/skills as first-class workflow primitives.

• Users want manager/worker patterns and pluggable agent capabilities.
• Evidence: Cursor requests for agent plugins/sub-agents, HN discussion mentioning manager/worker AI agent patterns.

5. Better approval/review ergonomics and safe autonomy controls.

• Users want granular approval policies (read-only, auto-accept for safe changes, review UX improvements).
• Evidence: Codex issues around approval modes, diff/review UX, and Cursor requests such as "Yolo mode for read-only commands" and review flow controls.

6. Cost/usage transparency and fairness on failures.

• Users care about usage predictability, especially when tasks fail.
• Evidence: Windsurf top-voted posts on credit refunds/rollover and failed-run charging frustration.

Recommended v1.0.2 Feature Set

P0 (Ship in v1.0.2)

[x] 1. Live Development Commentary Stream

• Stream concise step-by-step commentary while coding (plan -> action -> verification -> next step).
• Separate commentary from code/actions in the protocol and UI.
• Why: highest trust/value gain, directly addresses silent-agent frustration.

[x] 2. Agent Anti-Stall + Auto-Recovery

• Detect "stuck" states (repeated tool calls, no-progress loops, timeouts).
• Auto-recover by: retry strategy, tool backoff, summarize-and-continue, or explicit recovery prompt.
• Show recovery state in commentary timeline.

[x] 3. Execution Transparency Panel

• Show: selected model, tool invocations, active step, elapsed time, and token/cost estimate.
• Include explicit "why this action" snippets for each major step.

[x] 4. Safer Autonomy Modes

• Add clear execution modes: read-only, review-required, auto-apply-safe, full-auto.
• Add per-tool approval policy controls and visible current mode in UI.

[x] 5. Reliability Guardrails for Web/Tool Tasks

• Standardize tool schemas for strict providers.
• Add robust URL/tool validation and user-facing fallback messages.
• Add regression tests for schema compatibility across major model families.

P1 (Start in v1.0.2, finish in v1.0.3 if needed)

[x] 6. Persistent Project Memory (Scoped)

• Lightweight persistent project summary + architecture notes that survive sessions.
• Used to prime future runs and reduce repeated context rebuilding.

[x] 7. Sub-Agent Framework (Minimal)

• Introduce a simple manager/worker abstraction for long tasks (planner agent + executor agent).
• Keep initial version behind a feature flag.

Tool Schema Compatibility Matrix (v1.0.2)

Provider | Strict Tool Schema | Sample Models | Result -- | -- | -- | -- OpenAI | Yes | gpt-5-codex, codex-mini-latest, gpt-4o | Pass Anthropic | No | claude-3-5-sonnet-latest | Pass OpenRouter | No | openrouter/auto | Pass Together | No | meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo | Pass

Matrix data and live checks are exposed at:

• /api/system/tool-schema-matrix

Acceptance Instrumentation (v1.0.2)

Instrumentation now emits and tracks:

• commentaryFirstEventLatencyMs
• recoverySuccessRate (aggregate)
• manualInterventionRate (aggregate)

Available via:

• stream data event: run-metrics
• system endpoint: /api/system/agent-metrics

Suggested Acceptance Criteria

1. Commentary appears within 2 seconds of run start and updates at each meaningful step.
2. At least 90% of interrupted/stalled runs either recover automatically or produce a clear recovery suggestion.
3. Users can switch autonomy mode before and during a run; mode is reflected in every tool action.
4. Tool/schema compatibility tests pass for strict OpenAI Codex-style models and standard chat-completions models.
5. Timeline view logs model + tools + approvals + outcome for each run.

Proposed v1.0.2 Success Metrics

• Reduce "agent got stuck" reports by at least 40%.
• Increase successful end-to-end task completion rate by at least 20%.
• Reduce average "manual continue" interventions per session.
• Increase user satisfaction on "I understand what the agent is doing".

Sources (Playwright Browsed)

• https://forum.cursor.com/c/ideas
• https://forum.cursor.com/c/ideas/7
• https://codeium.canny.io/feature-requests
• https://news.ycombinator.com/item?id=44031432
• https://news.ycombinator.com/item?id=45534880
• https://github.com/openai/codex/issues
• https://github.com/features/copilot
• https://github.com/orgs/community/discussions/categories/copilot
• https://developers.openai.com/codex/

[embire2]

@embire2 have these features been added or should i add them or are we gonna add feature each of assigned a different feature

Drop me an email D.

[embire2]

We have just dropped version 1.0.3 of the new bolt.diy alternative, bolt.gives.

Have a look here: https://github.com/embire2/bolt.gives

We have done away with the bloatware and have clear roadmaps for future development.

We are busy working on a service where you will be able to, at no cost, create a mini instance of the software without touching a line of code.

We have been marketing the new builder on Reddit and have around 1000+ users using it actively per day now.

We hope you enjoy. Star us on Github and post any feature requests under Issues. I have a dev team of 20 x working on it.

[DevFlex-AI]

@embire2 do I count as dev team member because I coded some of it?

[embire2]

@embire2 do I count as dev team member because I coded some of it?

Absolutely! Your work is superb.

[DevFlex-AI]

@embire2 how to get in contact with the other devs

[embire2]

We have now rewritten the entire system from scratch. Join us in Alpha Mode here:

https://webcoder.codes

Alpha Free gets 100 x compute units per day. For more units, you can request with justification from the app.

We hope you enjoy.

[DevFlex-AI]

@embire2 is there a repo for this because I want to work on this project and be part of the dev team

[embire2]

Hi @DevFlex-AI ,

https://github.com/embire2/bolt.gives is still alive and is the open source model that our software company is working on. This contains bolt.diy as its main backend and is receiving a big update, version 3.0 today.

[embire2]

Version 3.0.1 of bolt.gives is dropping today. You can have a look here:

https://github.com/embire2/bolt.gives

It will include a Free provider, within reason, that will allow you to code at no cost. Dropdown the provider to FREE and the model will auto populate